“I. - -'—-- ‘v- I. LIV...‘ w"... (“fl—fir. - 1 ,,.. 4.... 0 -~ u- can '4:- o~ -.~‘/ _ -- 3.2..-.— - . 5 55mg” #45 m r....1mW4} 5 . :1 555555“ '5' 55"“‘5'1‘5' 55:55:51.5 "WI-J: 4!! 5"”:5 «Expmq'9' JV€:J h: 5W . imfi5 ' H556“: :‘5l’55155‘55 15555; ”'5 555:5? 55'5' _ :5 ' 5'55 ""55 53"” ‘ : :5” " 5'" ‘55‘ :5“: *‘k‘ 5 5: L: :2. 1:55:95!" :5 5'5 5 ' is": 5' 59%;}ka _. '5 “ .‘ a'gjk: p 3:}: :1 1:1: 13%;, '55; 4.21:: 15:55: '1': 9? 559+ g :5; 1:35;“, “‘55.,“ gr: ”5551 ,1 i .5!‘ -i‘3 5" ”5'5 . Vi '5 5" "A 5' 5" ‘ “05:14- '5'5 "5'555155'555'5'5'555'5‘ 55 v.ll"5::§_':l:(fl 5 1' ‘ ‘31 was 0—.-a“ m w . ' w}?- . . i to I 0‘ . 5 '(H ' :- 5'2"?“ "5‘1” [1, . : 5‘35: 1:: “15'5” ‘ :. 5-5 3"},5 55 "VP19‘*455(1“f‘:1§5525'5521%. {. . u5 y”: '55 4 5:: “H... ”13:45.5 1 algae-5 :. :555u4:5‘; NEW ”55:55:: ::' ‘ :5“: .r‘;5 5595‘“ ‘31:; ‘5: i: ‘53:}: 5‘5 51:}:5555 : >51 {a .‘ - : 555:5 “3" “(‘55“): gr. ”.gga 54 #54:: 5:555“ 555:} 5'5. 5'55 [#54: J5: ’4'45' 5'55 _: 5535' :2: '5 5"" '5'“ 55"5r5‘ ‘52,,“ l“ 5:5:5'. ‘5?”5315555'?” {3555555151 5‘; ' I 1'. ‘ '5'51‘55": ”‘ "U": -,., 55;;‘3 '4145.‘ 112...? :4“ ‘ - :3: '5555 a' :55: 55555653195555“ 555555;“: ‘. “‘5 E; 5 5' II [55-55% .55555 5' :5. .Ui “yum“: )5 5:55: “firth £5::5:'}}5:iu’nilfiff.5:. , E Egg“; 11:?“ 2‘ ‘i 3"! hi ”"1 Lg I :15; :23): 55:: 5' 'I" ::\I5 '55:! :1’, :4": "n: i" L: .. :3: 5155555535555: {:5 ”55"!“ - 'l:555455555:155‘5‘:55":' . K: 5158'}: kr'555 it: EYESfl 221:3.“ . '5; "'5' 55 5:55 5'55 "5““: 5 .. “'5“ 5:: 55%": “:35 5.‘:'v:‘5::‘1:‘:::=:*:"55'fi‘:“:5i::5::":?:::.‘::j.;;~..:,: :3: M :w: ‘ 51:“ . :::5:51‘2:5:::5‘I:z::;:: 5:55 555 '55 1'5: 5.: 5152:.5.5‘:5:'= “555::"G‘g5': 7255':57;;¥'5':5‘::‘:ié‘:::5;5:5 “55:55:55 1’55": :35". 5'" :5: :5“ 5:2 7:: :Ha : 1'5 H): . Mr} 55‘ .m: #5555 .1: 5:” r (:1: “:2“: :9 ~ 52.55: 5'. : 24%;; ‘4'! 5E .1: ”'5 u- t} 4N: 4.5231: 5-: 53:52'51155‘1511::5';55E?:::7"';"5,:55 {'5‘ "55“ "H 5.5.: ".43 ' '5‘. 55‘: 5'5155' .w“ £55 ”.13?- 55: : 5 ':"5"“’155§5::5::5‘8'5‘:'.5::.:‘5)5.5:51‘.~:;3:~‘55:5553::5':€p211.:i5::'::1“an":5.::§f'%u§55::55%55555'1 -: :3 5:55:55? :55‘1"‘5':5:U: :5" " :22: 5 2554 1' . ' Y '2 ' i ' 4.251 . H "21:: ' ' 5'5.“ 5. y ‘5':‘ \ . ' ‘h n I'sb‘v ' "4 :::::::5:53:,.:H:: “15:31“: i :5: :1: 5555555'5'55555'55'55'55‘5555‘455‘555’: "5"{5"‘555?'5'i:» 552;}:éL,W::§:§4:‘, f): 5:51.53 ”51‘5'592 41:1 51 . pm 0 ' M] . ‘5" ". 3:" ': “mg-z 501' ' Q; "I,“ 0:": «3.5:; “:J: fi::5:.553-::: j, :5 :55'55 ”5.. 5::‘:::::5::::::‘::::‘:::5:: 2:55,,“ [1:559 § 935': :95: F": :5.555"§1%5:'5'%53."“%:film: .-:{ 2:..‘5: :5 5‘:'E:I.?5'5.§:::::‘-:‘ 5:385" 551' ”5515.55 :; (3:; " "(-5 :‘::.5 555 5:15:55: H:2: .9 5"55‘ "5‘ 3:4“ 3:55}- :‘5' gem: :53 :-;"'5555 :55 : - '5'5';5:5:1‘55532351‘“! 5555'5'5'5'55‘5535 '34:“ "5" 1:41:35: .5; : '5'] 1 :IPJ 5:;5 0 v 4531:: {$11} 33%; 5.5.5". :5313‘“ 2.151515 .i. ‘ f ' ‘ “5:55; 4531535515.“555‘a:%:5‘-:5 55:25:56“:""‘ 5:33 55:: :-‘.‘3::v‘!-.' ":- “5’ .2: 153‘- 355:: ‘5 "2:5 ":2“: 1‘:“""""' ::23::::5:::5:::555:.2=:‘::::‘;4:222:52: 55:55.5. El; :15): :-.: =; 4 5.231521: “‘2“ 5‘55 '35 555:” ‘ W "M 1:2:::2:=:2=::~..-.;.«:.:=:.:::.: may“: «:2 . .. :2: ‘ :2: .|§:=:‘:i::52};:3"!‘11515535::1.;:‘5.5:5:::55:‘;5::'£‘:',:51 13,: ""51 5555:5951." 4:5,: ””5 ‘git‘: 2:: - 3:43,: .5.'5;:;,f.,?.:: -" '- 3- "i".':5':2':5:5il5:5‘.1 3:“:th :13! ' 23:54:59: 1:19.13? . 5‘52" “5%.. 31;]: .- L! 7% is :1 . "55" 5 ”5;: 33:3: 3:155: . fifiJHW‘5HMmM' 3 "5"5H5'5t'555”*”5'fiJ%?W55%5ww'i“" .- I' is" 4. ' '-:. v 5‘. 5. .255 . ‘ '53," ' ' " F 0 0 ,1" "I l - 5551415555555 "555; : L't:':'55555 "3.: 3"" : 2 i; 5.5.1 '555'5'5‘5 5"5 1'" 5'5"?" ‘ "r” 5 :5':}5:51:3‘:..5-,:. ‘ ;' . ::v..5....s' ‘ W "p “2:5: 25:?” -:r= ;v‘1;:‘55§'x‘ -‘ " 5 :5" 331:: :11. g" " 15.22:: . :g::. 5 "(95535: .' ' ‘ T'iW’ 9 ing, ’ "4:15.; 43.5%? v .» E" .2: t- :th’v: ‘ gv. F1:"':"‘"1';.27 2“: “5355.; '; . . :3;le '3 "5:55515'33'5‘155 "1 55:5 :2»: .- ... .:.».,...:.. :5 . 'uVWhp“fi55~u. .5w~flfi35"1_q: 555555'5'55'555":5:5"5"':':5‘ ’ ' 5"" ‘3 9'5: 55 i .‘5 5’51. ‘3 :ip', ,,. ...:.. .. _~=s :5" .. - , -:- , "'55. ‘ ‘ .5‘Y “2:21: .3':"'"~,1'1.sa-:'55' 405 '1.” 0 33:! ; N .' , :33 ,: .. 4M 3 _ . u 5'5" fl' ’ 1 ~:5§‘ 3% ‘:HJ'“ ,5" ‘ 51“ 0,::.,.';:', -, ;_‘ #3?“ .I_‘ ”5:515: In: 5'. '3' ‘55'5 . “ 1:” 5'.~"""555;"5'5 '. 57: p ‘ ..E:;:.::[_. ‘1‘“; l v“ U . s .5 ' 1' . J"... I. & f I '- .I'; , ' ' " :5: 5: 5"" :5'2555‘55‘" ': ~27f'3'5 4 II ‘45” f “’1‘, 5 3 Akin“. “Ali!“ “'25 :1 “5"} '515‘5'5' 55'n50555 1144‘! 55::5‘ 0:; 3::5' 3'5}: :5:f;':§;§'[:5:f. : . 0 ..... “ : 0:54.31 . g ' mu»: ."=2":=:.5.2::: . . '5 .5 :‘5 ' 52:5“. ll '2" 15:; I. 4"!" . . : L :1 3:4; .' ""5 55" 5" '5'"'5555'55::3525'555:5522?r : .2? :i’; ":5 u -1». .v—vy—t.‘ 551’- l I | ::J '5! J. “y? ' V'v'hll‘ 3.4: :1 I." '5'5615’55'5'5'5555'5: 1080:“ ‘ {[5 5&51’p 0 55555 '1' 1555'” “SQ“: . H: 0.. o--.»«-y—--—-c - 't .....--.....- - ,u. .4... — , - -v- 4 0-”... av" 0‘ PT": '55"; It" ,‘ 'i':5lt ." 4 : 3'"""55"5""""":'4. 1555! ::::::5l.::i:.:.'5:‘: 5515555555: 403:, 51:35 62'5" 2555,55 5:23:55”? 55“" '. -A. y. - .‘or‘ ‘ -s m» .— ”l-« .v. u- m": . ....o._-—-auq ‘W “Ca :45“ :5 f u"! ‘5 ’5‘ '5'! 5'55; :31? :5’13; ' 5 if; #59." 5%; l i .3|.5«1’.5::.""' ' 55': 5:355. 22325555555335:- .._... 3- ,_..._._.. ... m...— . .3- ‘Um- vw._ .. a"..-__.- . —‘W*-— - '1". 1: :1 :4. d 'n 4.-- ‘ o ...<.._. -.-‘ M n...“ ICHIGAN STATE U VE LIB S L ’ lull “1|le iwill:l‘llrlml'illll 3 1 93 01572 2097 This is to certify that the dissertation entitled Hardiness, Life Stress, and Neuroticism: A Structural Equation Model of Self-reported Illness presented by Lois A. Benishek l has been accepted towards fulfillment of the requirements for Ph.D. Counseling Psychology degree in Major professor V & Date 2-11-93 MS U i: an Affirmative Action/Equal Opportunity Institution 0-1277] LIBRARY Mlchlgan State University PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES mum on or More data duo. DATE DUE DATE DUE DATE DUE 24 U'I F 1 100802 lAMéUZfl ‘ * An ‘l A f r I.- usu 1.... .. .. + HARDINESS, LIFE STRESS, AND NEUROTICISM: A STRUCTURAL EQUATION MODEL OF SELF-REPORTED ILLNESS BY Lois A. Benishek A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Counseling, Educational Psychology and Special Education 1992 ABSTRACT HARDINESS, LIFE STRESS, AND NEUROTICISM: A STRUCTURAL EQUATION MODEL OF SELF-REPORTED ILLNESS BY Lois A. Benishek Kobasa's hardiness theory posits that persons exhibiting the personality characteristics of commitment, control, and challenge are less likely to report physical illness when encountering stressful life events. The hardiness construct is confounded with neuroticism and subjective illness reports. Recent studies have also begun to identify gender differences in how hardiness is expressed. The purpose of this study was to confirm the factor structure underlying two measures of hardiness, to evaluate Kobasa's theory when addressing recent criticisms of hardiness, and to investigate possible gender differences. One hundred and eighty-five university employees completed measures of hardiness, life stress, neuroticism, self-reported illness, and a more objective measure of illness behaviors. Confirmatory factor analyses did not identify the hypothesized three component model of hardiness. Further exploration of the three-factor model using a principal components analyses identified a five- factor solution underlying the Personal Views Survey and a two-factor solution underlying the Revised Hardiness Scale. Results from the structural equation models based on both ii frequency and severity scores identified differences in how hardiness is expressed in men and women; the models, however, were structurally weak. Implications for future research and practice are discussed. ACKNOWLEDGEMENTS The process of completing this research project was made significantly more enjoyable thanks to the contributions of several individuals. Fred Lopez provided me with an optimum balance of support and challenge and, most importantly, was readily available to me throughout my numerous "crises." His melding of a humanness and professionalism has provided me with a model I hope to emulate in my own professional career. I am grateful to Gloria Smith for supporting me throughout my master's and doctoral programs. I have an especially high regard for how she continually placed my professional interests before her own. Jeff Jensen was the bright spot throughout my SPSS and LISREL nightmares. He supplied me with excellent statistical guidance and constantly reminded me (through his expertise, sense of humor, enthusiasm, and bottomless jar of peppermints) that statistics is enjoyable. I thank Bertram Stoffelmayr for the seminal role he played in my professional development. Bertram provided me with a wide array of opportunities to learn about research, politics, the interplay between the two, as well as how to make a wicked cup of coffee. Aaron Jackson and Randi Kim provided me with ongoing emotional support and encouragement to "move on" when I became too compulsive about my analyses. Mary Anderson made it possible and painless for me to collect data in Michigan while I was living in Utah. Much more important than her contribution to my data collection efforts was the friendship and music she provided me when I was feeling frustrated, tired, and impoverished. Last but not least, I would like to thank Marilyn Benishek for teaching me the importance of hard work and dedication. iv TABLE OF CONTENTS LIST OF TABLES . . . . . . . . . . . LIST OF FIGURES . . . . . . . . Introduction . . . . . . . . . . . Problem Statement . . . . . . . . . . Review of the Literature . . The Relationship Between Life Stress and Health Psychological Hardiness: Kobasa' 5 Initial Research Kobasa' 5 Subsequent Research on Hardiness The Hardiness Buffering Effect The Factor Structure Underlying the Hardiness Construct . . . . . . . . . Principal components analyses Predictive strength of the hardiness components . . The Potential Confound Between Hardiness and Neuroticism . . . . The Potential Confound Between Neuroticism, Life Stress, and Illness . . . . . . . . . . . . Confound with life stress . . . . . . . . . Confound with illness . . . . . . . . . . . Sex Differences . Criticisms of the Hardiness Construct and Hardiness Research . . . . . . . . Measurement issues Inappropriate use of statistical methods Summary . . . . . . . . . . . . . . . Hypothesized Model . . . . . . . . . . . . . . . Methodology . . . . . . . . . . . . . . . . . . . . Subjects . . . . . . . . . . . . . . . . . . . Procedures . . . . . . . . . . . . . . . . . . Instruments . . . . . . . . . . . . . . . . . Hardiness . . . . . . . . . . . Personal Views Survey . . . . . . . . Revised Hardiness Scale . . . . . . . Stressful Life Events . . . . . . . Combined Hassles and Uplifts Scale . . PERI Life Events Scale . . . . . . . . viii U'lI-J mm Neuroticism . . . . . . . . 62 Neuroticism Scale from the NBC Personality Inventory . . . . . . . 62 Physical Illness Measures . . . . 63 Seriousness of Illness Rating Scale . . 63 Illness Behaviors . . . 64 Demographic and Background Information Form . 64 Research Hypotheses . . . . . . . . . . 65 Confirmatory factor analysis . . . . . . . . 65 Causal model . . . . . . . . . . . . . . 65 Data Analysis . . . . . . . . . . . . . . . . . . 66 Results . . . . . . . . . . . . . . . . . . . . 69 Descriptive Statistics . . . . . . . . . . . . . . 69 Inferential Statistics . . . . . . . . . . . 81 Confirmatory factor analyses . . . . . . . . 81 Personal Views Survey . . . . . . . . . 82 Revised Hardiness Scale . . . . . . . . 91 Principal components analyses . . . . . . . . 99 Revised Hardiness Scale . . . . . . . 103 Structural Equation Models with Latent Variables . . . . 114 Model for the whole sample based on. frequency scores . . . . 125 Model for men based on frequency scores 128 Model for women based on frequency scores . . 131 Model for the whole sample based on severity scores . . . 137 Model for men based on severity. scores 139 Model for women based on severity scores . . . . . . . . . . . . .. 143 Discussion . . . . 150 Relationships Among the Hardiness Variables . . 152 Unidimensional Versus Multidimensional Construct 153 Relationships Among the Variables Underlying Hardiness Theory . . . . . . . . . 155 The Neuroticism Confound . . . . . . . . . . . . 157 Structural Equation Modeling . . . 159 Model for whole sample based on frequency scores . . . 159 Model for men based on frequency scores . . 160 Model for women based on frequency scores . 160 Model for whole sample based on severity scores . . . . . 161 Model for men based on severity scores . . 161 Model for women based on severity scores . 162 Conclusions derived from structural models 162 Strengths and Limitations . . . 16S Implications for Future Research and Practice . 166 vi APPENDICES . . . . . . . . . . . . . . . . . . . . . 170 APPENDIX A . . . . . . . . . . . 171 Initial Contact Letter . . . . . . . . . . 171 APPENDIX B . . . . . . . . . . . . . . 172 Informed Consent Form . . . . . . . . . . . 172 APPENDIX C . . . . . . . . . . 173 Postcard Follow— Up Contact . . . . . . . . 173 APPENDIX D . . . . . . . . . . . 174 Second Follow- Up. Letter . . . . . . . . . . 174 APPENDIX E . . . . . . . . . . . 175 Third Follow- Up Letter . . . . . . . . . . 175 APPENDIX F . . . . . . . . . . . . . . 176 Personal Views Survey . . . . . . . . . . . 176 APPENDIX G . . . . . . . . . . . . . 179 Revised Hardiness Scale . . . . . . . . . . 179 APPENDIX H . . . . . . . 182 Combined Hassles & Uplifts Scale . . . . . 182 APPENDIX I . . . . . . . . . . 185 PERI Life Events Scale . . . . . . . . . . 185 APPENDIX J . . . . 189 Neuroticism Scale of the NEO. Personality Inventory . . . . . . . . . . . . . . 189 APPENDIX K . . . . . . 191 Seriousness of Illness Rating Scale . . . . 191 APPENDIX L . . 193 Demographic and Background Information Form 193 H bl uam qcn m prNIAr m 10 11 12 13 14 15 16 17 18 19 20 21 22 LIST OF TABLES Sample Demographic Information. . Sample Employment— Related Information Descriptive Statistics for All Variables. Correlations Among Measures in Structural Equation Model - Frequency . Correlations Among Measures in Structural Equation Model - Severity. Correlations Between Observed Variables Correlation Matrix Among Hardiness Variables. Internal Consistency of Scales. . Factor Pattern Results of the Confirmatory Factor Analyses for the Personal Views Survey . Goodness- of- Fit Indices for Confirmatory Factor Analyses. . Respecification Steps in Model- fitting Process for PVS. Factor Pattern Results of the Confirmatory Factor Analyses for the Revised Hardiness Scale. Respecification Steps in Model- fitting Process for RHS. . . Component Loadings for the Personal Views Survey Component Loadings for the Revised Hardiness Scale. . Box' 5 M Test for Gender Based on Frequency Scores . . . . . Box' s M Test for Gender Based on Severity Scores. . Univariate ANOVA by Gender - Frequency Scores . . . Univariate ANOVA by Gender - Severity Scores . . . Summary of Model Fit Information -- Whole Sample Based on Frequency Scores . . Summary of Model Fit Information -- Men Based on Frequency Scores. . . Summary of Model Fit Information —— Women Based on Frequency Scores. . . . . . viii 78 79 84 88 93 94 99 105 111 123 123 124 125 127 131 134 23 24 25 Summary of Model Fit Information -- Whole Sample Based on Severity Scores. . Summary of Model Fit Information -- Men Based on Severity Scores . Summary of Model Fit Information -- Women Based on Severity Scores ix 139 142 146 LIST OF FIGURES Figure Page 1 Proposed Structural Model — Composite Scores . . . . . . . . . . . . . 52 2 Proposed Structural Model - Component Scores . . . . . . . . . . . . 53 3 Revised Structural Model - Composite Scores . . . . . . . . . . . . 118 4 Revised Structural Model - No Measurement Error. . . . . . . . . . . . . 121 5 Final Structural Model - Whole Sample - Frequency. . . . . . . . . . . . . 130 6 Final Structural Model - Men - Frequency Scores . . . . . . 133 7 Final Structural Model - Women - Frequency Scores . . . . 137 8 Final Structural Model - Whole Sample - Severity . . . . . 141 9 Final Structural Model - Men - Severity Scores . . . . . . . . . . . . . 145 10 Final Structural Model - Women - Severity Scores . . . . . . . . . . . . . . . . . 149 CHAPTER I Introduction Psychologists have had a longstanding interest in personality factors that mediate adjustment to life stress (e.g., Cohen & Edwards, 1989; Contrada, Leventhal, & O'Leary, 1991; Holroyd & Coyne, 1987; Suls & Rittenhouse, 1987). Interest in this topic was motivated by the inability of physiological theories to consistently explain the effects of life stress (see Contrada et al., 1991 for a review). It was also prompted by evidence which suggested that, even when experiencing similar life changes, not all individuals exhibit illness (Hinkle, 1974). Hardiness is one personality characteristic proposed to play an important intervening role in the stress-illness relationship (Maddi & Kobasa, 1984). Hardiness theory was developed, in part, as a result of Kobasa's discontent with the overemphasis on unsuccessful coping processes, rather than successful coping processes, in the stress resistance literature. Kobasa noted that although some persons fall victim to the effects of life stress, others appear to benefit from these experiences. It is the latter group of persons which she referred to as stress resistent or hardy persons. 2 Kobasa's hardiness theory is theoretically grounded in existentialism. One of its basic premises is that an individual's perceptions of and actions in the world play an important part in shaping personality. A second premise is that life situations are always changing; this change provides opportunities for development and growth. Kobasa substantiated her theory through the research ‘ program she developed while she was a graduate student at the University of Chicago. In a sample of mid- to upper- level male business executives, Kobasa identified a group of high stress/low illness (hardy) and high stress/high illness' (non—hardy) persons. Hardy executives were those who reported greater levels of commitment, control, and challenge. Commitment is conceptualized as the ability to value oneself and one's life activities, control consists of the perception that one plays an instrumental role in influencing life events, and challenge involves the self- perception that change is a normal part of life and, therefore, is an opportunity for personal growth. The hardiness personality construct has been studied extensively in the past decade. It has been consistently associated with lower levels of self-reported illness in people reporting highly stressful lives (Kobasa, Maddi, & Courington, 1981; Kobasa, maddi, & Kahn, 1982; Kobasa, Maddi, & Puccetti, 1982; Kobasa & Puccetti, 1983). Research has identified additional personality and life functioning characteristics which either interact with hardiness or have 3 an additive effect in decreasing illness reports (Contrada, 1989; Ganellan & Blaney, 1984; Kobasa, Maddi, Puccetti, & Zola, 1985. Hardiness has also been associated with other types of adjustment such as lower levels of depression (Ganellan & Blaney, 1984; Rhodewalt & Zone, 1989) and occupational burnout (Nowack, 1986). According to Kobasa, cognitive processes associated with hardiness can be learned, and worksite wellness programs have been developed to promote these characteristics in employees (Maddi & Kobasa, 1984). Although these findings have promising implications for health promotion as well as life and work satisfaction, hardiness has recently been criticized on a number of conceptual and methodological grounds. At the conceptual level, critics have questioned the actual number of components underlying hardiness. Although the majority of principal components analyses have identified three components, other studies have identified as few as two and as many as four components. These analyses have been conducted primarily on males, college students, and other homogeneous groups. Second, hardiness has been typically studied using negative indicators. Critics question the validity of research findings that are based on the assumption that it is possible to measure the true opposite of hardiness. Related to this is a third criticism: hardiness may simply reflect the absence of neuroticism. Correlational studies indicate that these two constructs are 4 strongly correlated but not to the extent that they are identical. Controlling for the effects of neuroticism significantly changes the effects of hardiness on outcome variables. The effects often decrease in magnitude or disappear entirely. There is also evidence that neuroticism may be confounded with life stress and illness-related variables. Fourth, hardiness was originally validated on a sample of men. Although some evidence in support of sex differences is emerging from the literature, little is know about the similarities and differences in how hardiness affects life stress and illness in men and women. Finally, there is inconsistent evidence that hardiness acts as a buffer against the negative effects of life stress. Numerous methodological criticisms have also recently been raised about hardiness. The first of these criticisms is the tendency of researchers to study composite scores and overlook the possible individual contributions each of the three hypothesized components has on life stress and illness. Second, a variety of questionnaires have been developed to assess hardiness. Limited attention has been given to the psychometric properties of these tools. Third, inappropriate statistical techniques have been used to analyze hardiness effects and hardiness-life stress interactions. These techniques often fail to control for possible confounds or they treat hardiness as a dichotomous rather than a continuous variable. Pr 1 m S a m nt There is a need for a more systematic evaluation of existing hardiness theory, its measures, and the relationship between hardiness and other constructs such as neuroticism. Conceptual and empirical clarification is a necessary first step toward understanding the relationship among hardiness, life stress, and neuroticism and how these variables influence physical illness. The purposes of this study are to clarify a) whether hardiness is a unidimensional or a multidimensional construct, b) differences in the strength of the relationship between the hardiness composite, its components, life stress, and different measures of illness when accounting for the effects of neuroticism, and c) the possible presence of gender differences in the hardiness- life stress paradigm. I Should support be found for the hardiness research paradigm, greater attention can be given to developing hardiness-promotion programs. A lack of support for the hardiness paradigm would suggest that other personality variables, such as neuroticism, should be examined more closely in order to understand why certain people report greater physical illness than others who are experiencing similar degrees of life stress. CHAPTER II Review of the Literature The Relationship Between Life Strese and Health Professionals in the fields of medicine and psychology have had a longstanding interest in the ability of personality factors to mediate adjustment to life stress (e.g., Cohen & Edwards, 1989; Contrada et al., 1991; Holroyd & Coyne, 1987; Suls & Rittenhouse, 1987). Hippocrates and Galen were among the earliest persons reported to have an interest in the link between personality and illness (see Contrada et al., 1991 for a review). Their conceptualization focused on the relationship between bodily fluids and personality types. These biopsychological characteristics were linked with the tendency to experience certain types of illnesses. More recently, Selye's general adaptation syndrome (1956) drew attention to the notion that stressful life events can accumulate to the extent that the organism becomes exhausted. This exhaustion can manifest itself in a variety of illnesses and even death. Holmes and Rahe (1967) suggested that life events which require adjustment in a person's daily routine were stressful. As a result of this stress, people were more likely to become ill. 6 7 Absent from much of this early work was the acknowledgement that a significant minority of people lead very stressful lives and yet do not report high levels of illness. That is, some people appear to be more resistant to the effects of life stress than others. Prior research indicates that there is a small, yet reliable relationship between life stress and illness symptoms. Correlations average from about 0.20 to 0.40 (Kobasa et al., 1981; Rabkin & Struening, 1976a, 1976b; Roth, Wiebe, Fillingham, & Shay, 1989). The great variability among these scores (i.e., standard deviations have been as large as eight times the mean) suggests that similar degrees of life stress have substantially different effects on illness behaviors (Kobasa, 1982b; Maddi & Kobasa, 1981). These findings prompted researchers to examine more closely the personality factors that may mediate the stress—illness relationship. One such personality characteristic is hardiness. Psyeholggieel Herdiness: Kobeea'e Initiel Research Kobasa's doctoral dissertation provided the basic framework for understanding how one personality characteristic, hardiness, affects the stress-illness relationship (Kobasa 1979a; 1979b). According to Kobasa, hardy people (i.e., people leading highly stressful lives and yet not reporting physical illness symptoms) exhibit three cognitive coping strategies. These strategies or components of hardiness are referred to as commitment, 8 control, and challenge. Commitment is the tendency to believe in the value of what one does or to have a sense of purpose and meaningfulness in life's endeavors. Control is conceptualized as one's perceived ability to influence life events and to see oneself as influential rather than helpless. Challenge refers to the use of optimistic cognitive appraisal to perceive change rather than stability as being a normal part of life and as being beneficial to one's personal development. Six personality scales were hypothesized to measure each of the three hardiness components. These scales were selected from both established personality inventories and more recently developed personality assessment tools. Questionnaires containing these eighteen personality scales were mailed to 837 mid- to upper-level executives employed by Illinois Bell Telephone. Of this sample, women (n = 22) and low stress cases (n = 322) were discarded. As such, Kobasa's investigation was based on a high stress/low illness group (n = 86; i.e., hardy) and a high stress/high illness group (n = 75; i.e., nonhardy) of male executives only. Discriminant function analyses identified two scales for each component which differentiated between hardy and nonhardy executives. For commitment, these scales were Alienation from Self and Alienation from Work, for control, these scales were Locus of Control and Nihilism, and for challenge, these scales were Vegetativeness and 9 Adventurousness. Somewhat similar results were reported in a later study (Kobasa, Hilker, & Maddi, 1979). Kobasa's initial work suggested that hardiness "buffered" individuals against the development of illness. That is, persons leading stressful lives and exhibiting high levels of commitment, control, and challenge were less likely to report illness symptoms in comparison to similarly stressed people without the hardiness quality. Kobaee'e Subsequent Research on Hardiness Kobasa conducted a series of retrospective and prospective studies to further validate the hardiness construct and its buffering effect. Each of these studies explored the relationship between hardiness and other personality and life functioning variables. The majority of these studies were conducted on male mid— to upper-level managerial employees. It is important to note that different combinations of hardiness scales were included in Kobasa's assessment battery. With the exception of one study (i.e., Kobasa, 1982a), composite scores were used as the measure of hardiness. These studies are described briefly below in chronological order. Kobasa and her colleagues first published a five-year prospective study on the relationship between hardiness and constitutional predisposition (i.e., parents' illness reports) among male executives (Kobasa et al., 1981). The results from this study indicated that hardiness was associated with less illness and constitutional 10 predisposition was associated with more illness. After controlling for initial levels of illness, however, a relationship was not found between stressful life events and future illness. Hardiness buffering effects were not found in either analysis. A second study presented the results from concurrent and prospective analyses using initial levels of illness as a covariate (Kobasa et al., 1982a). In both analyses hardiness buffered against the effects of stress on illness. A significant main effect of hardiness on illness was also found. Stress had a direct effect on illness in the concurrent analysis but not in the prospective one. The role of hardiness and exercise (i.e., involvement and degree of strenuousness of sport and non-sport-related activity) among male executives was also studied (Kobasa et al., 1982b). Higher levels of stressful life events were associated with greater illness reports. Hardiness and exercise functioned independently to decrease illness. Both hardiness and exercise interacted with stressful life events, indicating that these variables are particularly important at minimizing illness as the level of stress increases. The relationship among commitment, coping, social support, fitness, and stressful life events was studied within a mixed sex sample of general practice lawyers (Kobasa, 1982a). Contrary to Kobasa's previous findings, no significant relationship between the level of stressful life .b' «A- cJ «‘5 -A. .u. q\~ .mu. eh. ‘I 11 events and the number of illnesses was found. Stressful life events were, however, predictive of strain (i.e., physical and mental symptoms typically associated with stressful life experiences). People who were either high in commitment or who did not use regressive coping strategies (i.e., efforts to deny, minimize or avoid stressful situations) reported less strain. Social support was only slightly predictive of strain, with greater levels of social support associated with more strain. Exercise was not a significant predictor of strain. Kobasa did not investigate whether there was a commitment-life stress buffering effect in this study. Kobasa investigated the relationships between hardiness, social support, and resistance to the physical effecms of stress among a sample of male executives (Kobasa & I”uccetti, 1983). Two separate ANOVAs were conducted; the firfiit analysis included a measure of boss support and the secCInd included a measure of family support. In the analqysis that included family support, hardiness had a dirfiact effect on illness, and life stress had a buffering effect on illness. This effect, however, was not identified in tlhe analysis that included boss support. Social assets (e‘SJ., parental occupation, father's educational level, exteant of group membership) were not significantly related to illness. Kobasa and her colleagues also examined the relJitionship between hardiness and Type A behavior among 12 male executives (Kobasa, Maddi, & Zola, 1983). Type A behavior is defined as a personality style in which people "display excess achievement striving, competitiveness, impatience, hostility, and vigorous speech and motor mannerisms" (Friedman & Rosenman, 1974). Hardiness and stressful life events were significantly related to illness reports. A work-stress by hardiness interaction effect indicated that people exhibiting Type A behavior pattern were more likely to report high levels of physical illness when they experienced high levels of stress levels and low levels of hardiness. Most recently, Kobasa and her colleagues conducted a concurrent and a prospective study on the effects of hardiness, exercise, and social support on illness among male executives (Kobasa et al., 1985). The purpose of the study was to determine whether a larger number of resistance resources would be associated with less illness. Both the concurrent and prospective analyses used in this study identified an inverse relationship between the number of resistance resources and the level of illness reported. The greater the number of resources available to people, the less likely they were to report physical illness. Findings ; from a multiple regression analysis indicated that hardiness, exercise, and social support (in descending order of strength) were significant predictors of less illness. Hardiness made an even larger contribution to the outcome variance when future illness was used as the dependent 13 measure. Exercise and social support, however, accounted for less of the outcome variance than they did in the concurrent analysis. No hardiness buffering effects were investigated in this study. In conclusion, six points of clarification should be made about Kobasa's research. First, Kobasa's findings are derived primarily from male professionals. Second, her findings are based on‘a variety of measures of hardiness. Unfortunately, rationales for the additions, deletions, and scale combinations used to assess hardiness were not consistently provided. Third, with one exception (i.e., ‘ Kobasa, 1982a), hardiness was always assessed using composite scores. Kobasa did not investigate the effect of each hardiness component on illness. Fourth, Kobasa used a single factor analysis and correlational evidence to support her three component theory of hardiness. Fifth, with two exceptions (i.e., Kobasa et al., 1981; Kobasa & Puccetti, 1983), the results from these studies are strongly supportive of hardiness's ability to act as a buffer against illness in stressful life circumstances. Sixth, only rarely did Kobasa examine the impact of other personality characteristics on hardiness's effect on illness. The EBIQIHBES Buffering Effeet By definition, hardiness acts as a buffer when it interacts with life stress to influence the dependent variable of interest. Support for the hardiness buffering effect is inconclusive. After presenting the literatures 14 that both support and refute the buffering effect, possible reasons for these contradictory findings will be discussed. Nine studies provided support for the hardiness buffering effect. Four of these studies were conducted by Kobasa (i.e., Kobasa et al., 1981; Kobasa et al., 1982a; Kobasa et al., 1982b; Kobasa & Puccetti, 1983). These studies all used self-reported illness as the dependent variable. In one of these studies, (i.e., Kobasa & Puccetti, 1983), buffering effects were found in an analysis which contained a measure of family support but not when boss support was included in the analysis. , In addition to self—reported illness (Rhodewalt & Zone, 1989; Roth et al., 1989), hardiness and life stress have been shown to buffer against depression (Ganellan & Blaney, 1984; Rhodewalt & Zone, 1989), occupational burnout (Nowack, 1986), and to be associated with positive self-statements (Allred & Smith, 1989). Results from an equal number of studies, however, do not find support for the hardiness buffering effect. Kobasa published two such studies using self—reported illness as the dependent variable (i.e., Kobasa et al., 1983; Kobasa & Puccetti, 1983). Other studies using illness as the outcome variable found a similar lack of support for the hardiness buffering effect (i.e., Funk & Houston, 1987; Rhodewalt & Zone, 1989; Schmied & Lawler, 1986; Wiebe & McCallum, 1986; Wiebe, Williams, & Smith, 1991). Similarly, studies using depression (Funk & Houston, 1987; Rhodewalt & Zone, 1989), 15 psychological distress (Nowack, 1986), and occupational burnout (Barry, 1988) as dependent variables failed to identify a hardiness buffering effect. The inconsistency of these results may be a result of several factors. First, much of the research uses a median split method to identify subgroups of people who are high or low on variables such as hardiness and life stress. This is not an appropriate test of the hardiness buffering effects, because this type of analysis tests for differences in the amount of variance explained (correlation coefficients) rather than for the difference between slopes (regression coefficients; Cohen & Edwards, 1989). Second, differences in the samples studied may contribute to the contradictory findings. Although male business executives (e.g., Kobasa's research) and students (e.g., Funk & Houston, 1987; Ganellan & Blaney, 1984; Wiebe & McCallum, 1986; Wiebe et al., 1991) are frequently studied groups, other samples such as human service workers (e.g., Nowack, 1986), female secretaries (Schmied & Lawler, 1986) and the elderly (Barry, 1988) have also been used to investigate the hardiness research paradigm. Third, the variety of assessment tools used to measure hardiness, life stress, and illness may produce inconsistent hardiness buffering effects. For example, six different measures of hardiness were used in these nineteen studies. Fourth, hardiness may not act as a buffer against all life functioning characteristics (e.g., self-reported illness, depression, occupational burnout) to the same 16 extent. Finally, buffering effects may be masked through the use of composite scores rather than component scores. Certain of the three hardiness components may interact with life stress to decrease its detrimental effects whereas others may not. The Feeeer Structure Underlying the Hardineee Construet There is a lack of consensus regarding both the number of components underlying the hardiness construct, as well as the predictive strength of each of the components. The results from nine factor analyses using five different measures of hardiness by six research teams will be presented. Following this, research addressing the predictive strength of each component will be presented. Prineipal components enalysee. Kobasa conducted a second-order principal components analysis on her six-scale measure of hardiness (reported in Kobasa et al., 1981; Kobasa, 1982b). Each of these scales consisted of negative indicators of the hardiness components. A personal communication with Kobasa (as cited in Hull, VanTrueren, & Virnelli, 1987) indicated that the analysis was conducted using an oblique rotation. Items with loadings greater than .30 on the extracted factors were retained. The subject pool consisted of male business executives. A general hardiness factor accounted for 46.5% of the explained variance. Each of the scales correlated .44 to .89 with the general factor with the exception of the Cognitive-Structure scale (r = -.01). The Cognitive- 17 Structure scale was the sole scale that loaded on the second factor. It accounted for 18.5% of the variance. Kobasa deleted this scale from the hardiness questionnaire since 1) it did not load significantly on the general factor and 2) a review of the item content suggested that it was not measuring her conceptualization of challenge. Kobasa's justified conceptualizing hardiness as a three factor construct based on her finding that the scales for each construct correlated more highly with themselves than they did with the scales associated with the other constructs. Formal efforts to substantiate this notion (i.e., completing first-order principal components analyses) were not completed. Hull and his colleagues conducted a total of three factor analyses on a sample of college students (Hull et al., 1987). Hardiness was assessed using Kobasa's original six-scale measure of hardiness as well as her Revised Hardiness Scale (RHS). A first-order principal components analysis was completed on the six-scale measure of hardiness using an oblique rotation. Items with loadings greater than .30 were retained in this analysis. The commitment, control, and challenge components were identified. Their eigenvalues were 8.93, 3.91, and 3.63, respectively. Collectively, they accounted for 18% of the explained variance. Four of the six scales loaded somewhat consistently on the hypothesized factors. Alienation from Self and 18 Alienation from Work loaded on commitment, External Control loaded on control, and Cognitive-Structure loaded on challenge. The remaining three scales did not load consistently on their hypothesized factors. Powerlessness loaded consistently on commitment rather than on control. Security loaded weakly on both commitment and control, and not on the hypothesized challenge component. Hull and his colleagues compared Kobasa's factor loadings with their own findings on the Revised Hardiness Scale. Of the thirty—six items, only twenty—five loaded on the hypothesized factors. Eleven of the twelve commitment, nine of the sixteen control, and four of the eight challenge items loaded as expected. As an extension of the same study, data from two college samples were used to conduct a pair of first-order principal components factor analyses on the Revised Hardiness Scale. Results from both samples identified the three hardiness components as commitment, control, and challenge. Eigenvalues from one sample were 4.68, 2.56, and 1.95 for each factor, respectively. The factors accounted for 26% of the variance. The eigenvalues identified from the second sample were 4.93, 2.21, and 2.14, respectively. This model also accounted for 26% of the explained variance. Similar to their earlier findings, not all of the items loaded on the hypothesized factors. Using data collected from male college students, Funk and Houston (1987) conducted a first-order principal 19 components analysis on Kobasa's five-scale hardiness measure. Contrary to the results of previous factor analyses, Funk and Houston identified a two-factor solution. The eigenvalues for the two factors were 2.36 and 1.06. Collectively, they accounted for 69% of the variance. Similar to Hull's research findings, the scales did not load consistently on the predicted components. The two measures of commitment (Alienation from Self; Alienation from Work) and one of the control measures (Powerlessness) loaded most strongly on the first factor. Security (a measure of challenge) and External Control (a measure of control) loaded on the second factor. McNeil, Kozma, Stones, and Hannah (1986) conducted two sets of principal components analyses on the 20-item Abridged Hardiness Scale. Data were obtained from people who were predominantly over sixty years of age. The two principal component analyses were separated by a one-year time interval. The initial pair of first-order principal components analyses identified three factors with eigenvalues equal to or greater than 1.5. These analyses accounted for 31% and 32% of the total variance, respectively. After completing a Varimax rotation, thirteen of the twenty items loaded greater than .40 on the three factors. Only ten of the thirteen items loaded as theory predicted (i.e., 3 commitment, 4 control, 3 challenge items). The authors interpreted these findings as being supportive of the three 20 component hardiness structure. Three of the seven mis— loaded items loaded as hypothesized in a second factor analysis which was completed on an independent data set. McNeil and his colleagues went on to conduct a pair of second-order principal components analyses to determine whether the subscales loaded on a single general factor. A single factor with an eigenvalue greater than 1.0 was obtained from the data collected at both time points. This general factor accounted for 49% and 47% of the variance, respectively. All three components loaded .45 or greater on the general hardiness factor. Morrissey and Hannah (1986) completed a principal components analysis on an adolescent version of the Abridged Hardiness Scale. After eliminating seven items because of their low item-total correlations, the analysis identified four factors. These factors were interpreted as control, challenge, commitment to school, and commitment to self. They accounted for 48.7% of the total variance. Each of these factors loaded greater than .50 on a single second- order factor. This general factor accounted for 40% of the variance. Pollock and Duffy (1990) developed their own unique measure of hardiness, the Health-Related Hardiness Scale. The item content was developed with the intention of assessing Kobasa's three components of hardiness. Ten of the original fifty-one items were deleted because of their low item-total correlations. A first—order principal 21 components analysis was then conducted on the remaining forty-one items using an oblique rotation. A two-factor solution was identified with thirty-four of the items loading .35 or greater on the hypothesized factors. The two factors accounted for 32.1% of the variance and had eigenvalues of 8.2 and 2.9. The first factor was identified as a combination of challenge and commitment, and the second was interpreted as control. A number of conclusions can be drawn from these principal components analyses. First, hardiness is a multidimensional construct which consists of at least two components. Second, the components are not measured equally well. Commitment is the most precisely measured component, followed by control and challenge. Third, six hardiness measures have been factor analyzed using data obtained from a variety of relatively homogeneous populations. The generalizability of these results is questionable. P d' iv r n th of the bar in m . The vast majority of hardiness research has been conducted using composite scores. Composites scores are calculated by combining standardized scores from the three equally- weighted component scores. The frequent use of composite scores may be a result of two considerations. First, Kobasa set a precedent for using composite scores with her own research. Others may have followed her procedure without questioning their potential limitations. Second, composite scores are appealing because they simplify the data analysis 22 and interpretive aspects of research (Carver, 1989). The benefit of the enhanced simplicity of using composite scores is tempered by a) the loss of explanatory information about each component, b) their inability to identify possible synergistic (i.e., interaction) effects among the components (Carver, 1989), c) their inability to allow comparisons to be made across samples studied, and d) their inability to develop normative information. Given that the principal components analyses have consistently identified at least two components underlying hardiness, questions can be raised about each component's ability to predict the outcome variable of interest. For example, does each component possess comparable predictive strength? Is their predictive ability similar across a variety of outcome variables (e.g., illness, depression, occupational burnout)? Unfortunately, little attention has been given to the independent roles that commitment, control, and challenge play in mediating the stress-illness relationship. The studies that have investigated the specific effects of the hardiness components on illness and other outcome variables are reviewed below. Kobasa herself published only one study in which she investigated the relationship between commitment, coping strategies, and illness-related variables among lawyers (Kobasa, 1982a). She found that lawyers who were more alienated and tended to use regressive rather than active coping styles in stressful situations were more likely to 23 report strain (i.e., physical symptoms typically associated with physical or mental overexertion). Manning and his colleagues reported basic correlational information regarding the components' relationship to a variety of health-related outcome measures (Manning, Williams, & Wolfe, 1988). Both commitment and control were consistently correlated in the expected direction with a variety of health and life stress variables whereas challenge was not. Schmied and Lawler (1986) explored the relationship between hardiness and its components, Type A behavior, life stress, and illness among a sample of female secretaries. Only the Powerlessness scale, a measure of control, was significantly correlated with the frequency of illness reported. When the hardiness variables were entered into a regression equation, however, neither the hardiness composite or any of the three components differentiated between high stress/high illness and high stress/low illness women. Holt, Fine, and Tollefson (1987) published a second study based on an exclusively female sample. They found that women scoring high on the commitment dimension were less likely to report a high number of stress-related illnesses. Roth and his colleagues examined the predictive effects of hardiness, life stress, and fitness on illness among college students (Roth et al., 1989). In comparison to the A AOfi’ ...a— ‘ .ubv or.- 5,. ~46- “a“ - -”".. Q.. p "O “... 24 other two components, commitment was the strongest predictor of illness. Higher levels of commitment were associated with fewer illnesses. Neither control nor challenge appeared to offer any significant health-related benefits. Contrada (1989) examined the relationship between the hardineSs components and cardiovascular functioning (i.e., diastolic blood pressure) among male college students. Only the challenge component was predictive of changes in blood pressure. One study examined the impact of the particular hardiness components and life stress on the physical and mental health of adolescents (Shepperd & Kashani, 1991). With regard to somatic complaints, adolescents who scored low on commitment and control were more likely to report illness. The health-related effects of hardiness among members of an agricultural organization were also recently reported (Lee, 1991). Only the control dimension of hardiness was a significant predictor of perceived physical health. Wiebe and her colleagues (Wiebe et al., 1991) investigated the predictive strength of each hardiness component on self-reported illness among college students. For their mixed-sex sample, challenge was a significant predictor of illness, whereas control's ability to predict illness only approached significance. The effect of commitment was insignificant. A different pattern of results emerged for men and women. These specific findings '1 1‘! ‘. l 4 an. no HA v9 Ill Uh .A ( (I) “I 25 are discussed in a subsequent section. Significant relationships between hardiness components and scale scores with outcome variables other than illness have also been reported in the literature. At the component level, commitment and control have been associated with certain attributional styles (Hull, VanTreuren, & Propsom, 1988; Hull et al., 1987), as well as with increased empathy, cooperation, and friendliness (Leak & Williams, 1989). Commitment has been shown to buffer against the onset of depression (Gill & Harris, 1991; Lee, 1991; Shepperd & Kashani, 1991) and job burnout (Holt et al., 1987). People scoring high on commitment are more likely to have an optimistic outlook on life, have more self-esteem and interest in social activities, and be more introspective (Hull et al., 1987). People reporting a greater degree of control are more likely to be optimistic, report higher levels of self-esteem (Hull et al., 1987), and are less likely to be depressed (Hull et al., 1987; Lee, 1991; Shepperd & Kashani, 1991). At the scale level, Alienation from Work is positively correlated depression (Funk & Houston, 1987) and occupational burnout (Keane, DuCette, & Adler, 1985). Low scores on the Alienation from Self and Vegetativeness scales are predictive of greater depression (Ganellen & Blaney, 1984). In summary, research supports the notion that hardiness is a multidimensional construct. However, the number of 26 components underlying the construct is not clear. There is also some indication that the hardiness components may have a differential effect on illness and other outcome variables. These findings suggest that the continued use of hardiness composite scores may limit the practical utility of the information derived from research on hardiness. One important avenue to pursue is that of investigating the unique contribution of each component on health-related variables. The Potentiel Confound Between Hardiness end Neureeieiem The following two sections present research suggesting that neuroticism is a potentially potent confound in the hardiness-illness research paradigm. Neuroticism, one of the five major dimensions of normal personality (Contrada et al., 1991; McCrae & Costa, 1987), is characterized as a tendency to view the world in a negative light (Costa & McCrae, 1987; Eysenck & Eysenck, 1964; Watson & Clark, 1984; Watson & Pennebaker, 1989). Persons high in neuroticism are "prone to experience fear, anger, sadness, and embarrassment; are unable to control cravings and urges; and feel unable to cope with stress" (Costa & McCrae, 1987, p. 301). The concern that hardiness and neuroticism may be confounded arises, in part, from measurement-related criticisms of hardiness for its use of negative indicators (Funk & Houston, 1987; Hull et al., 1987). Rather than assessing hardiness directly, early measures of hardiness 27 consisted of negative indicators of the construct. That is, the absence of hardiness was indicated by high scores on alienation from work and self, powerlessness, internal control, and the need for security. The actual content of the hardiness scales appears to be similar to that found in measures of neuroticism (e.g., Commitment: Life is empty and has no meaning; Control: Often I do not know my own mind). These similarities raise the issue that the presence of hardiness, in part, may reflect the absence of neuroticism (Allred & Smith, 1989; Funk & Houston, 1987; Rhodewalt & Zone, 1989; Wiebe et al., 1991). The relationship between hardiness and neuroticism has been investigated by a number of researchers. Some of these studies simply report the correlations between hardiness and neuroticism. Depending on the measures used, these correlations range from .24 to .62 (Allred & Smith, 1989; Hull et al., 1987; Massey, 1989; Rhodewalt & Zone, 1989; Wiebe et al., 1991). Others have addressed the issue of confounding more directly by examining both the strength and durability of the hardiness effects after controlling for initial levels of neuroticism. Some of these studies use the traditional dependent variable of self-reported illness. Others have investigated the relationship of hardiness to other psychological variables. Two research teams did not find differences in their results after controlling for neuroticism. Type A persons 28 continued to report high levels of psychological distress, and hardy persons reported less distress (Nowack, 1986). Allred and Smith (1989) found no difference in the number of positive self-statements reported after controlling for neuroticism. The significant effect for negative self- statements, however; disappeared once neuroticism was statistically controlled. Results from these studies counter findings elsewhere in which the hardiness effects either decreased in magnitude or were totally eliminated after neuroticism was controlled (e.g., Allred & Smith, 1989; Funk & Houston, 1987; Rhodewalt & Zone, 1989; Wiebe et al., 1991). Only the three studies that used self—reported illness as a dependent variable are reviewed below. Funk and Houston (1987) were the first to identify contradictory results depending on whether or not neuroticism was statistically controlled. Correlational analyses identified a significant relationship between hardiness and neuroticism among a sample of male introductory psychology students. The hardiness composite correlated .25 and .40 with two measures of maladjustment (i.e., neuroticism). Correlations between maladjustment and the individual hardiness scales ranged from .00 to .37. A series of ANOVAs and ANCOVAs, as well as multiple regression analyses were completed. Data were collected over an eight week period of time. The findings differed substantially according to whether or not the analysis was retrospective 29 or prospective in nature. In the retrospective analysis, the majority of the hardiness effects disappeared after controlling for neuroticism. Specifically, hardiness was no longer associated with differences in health problems, whereas the effect on depression remained significant. In the prospective design, the main effects for hardiness on subsequent depression remained significant regardless of whether or not neuroticism was controlled. No significant effects for hardiness on later illness were found using either ANOVA or ANCOVA. Finally, retrospective and prospective multiple regression analyses were completed. These results differed from both the ANOVA and ANCOVA findings. No main effects were found for hardiness on either health problems or depression when the retrospective data were used. As was the case with the ANCOVA, a main effect of hardiness on depression was identified in the prospective data analysis even after controlling for one measure of neuroticism. This hardiness effect only approached significance when a second measure of neuroticism was used as a covariate. Hardiness did not have a significant effect on illness. These basic findings were replicated in a study conducted by Rhodewalt and Zone (1989). After controlling for depression, neither hardiness nor life change events predicted illness. Another extensive evaluation of the potential confound between hardiness and neuroticism was conducted on a sample 30 of college students (Wiebe et al., 1991). Two measures of both hardiness and neuroticism were used. Neuroticism was significantly correlated with the hardiness composite and each composite score. Correlations ranged from .21 to .61 between hardiness and neuroticism. All correlations were in the expected direction. Results from a multi-trait monomethod analysis indicated that the control and challenge components were more confounded with neuroticism than were the hardiness composite and commitment. Multiple regression analyses were also completed using these measures in addition to a measure of life stress. The analyses indicated that the hardiness composite, commitment, and control were predictive of fewer illnesses when neuroticism was not statistically controlled. The PVS challenge component was not statistically significant, although the RHS challenge component was significant. After controlling for neuroticism, neither the composite score or any of the three components were predictive of illness for both the PVS and the RHS. In summary, the correlations among various measures of both hardiness and neuroticism clearly indicate that hardiness is confounded with neuroticism. As such, the relationship between hardiness, life stress, and self- reported illness may be a reflection of neuroticism rather than hardiness. The correlations are not strong enough, however, to suggest that hardiness and neuroticism are completely redundant constructs. 31 Results from more complex analyses, such as ANCOVA and multiple regression with a covariate, have attempted to clarify the relationship between hardiness and neuroticism. The effects of hardiness on illness reports tend to decrease or disappear when neuroticism is controlled. Rather than viewing hardy people as being particularly adept at overcoming the negative effects of life stress, it may be 7 more accurate to interpret these findings as indicating that non-hardy people are more psychologically maladjusted or neurotic than hardy people. The Potential Cenfound Between Neuroticism, Life Streee, end Illneee Psychosomatic research often links neuroticism (i.e., anxiety, depression) to disease (Dohrenwend & Dohrenwend, 1981). Historically, research findings on psychosomatic illness have indicated that emotionally distressed people report higher levels of life stress and illness than do non- distressed people. At first glance, the correlations between neuroticism and symptom reports appear to support this notion. For example, people experiencing greater emotional distress are more likely to report more medical symptoms (r =.44; Blazer & Houpt, 1979; Costa & McCrae, 1985a; Costa & McCrae, 1987). This phenomenon has been identified across a variety of populations (Tessler & Mechanic, 1978). The relationship between life stress and illness has been questioned recently on the grounds that statistically 32 significant research findings may be the result of a confound with neuroticism. That is, neuroticism may be confounded with the measurement of both life stress (Dohrenwend, Dohrenwend, Dodson, & Shrout, 1984) and illness reports (Costa & McCrae, 1987). As a result, research findings may not be accurately representing the relationship between hardiness, life stress, and illness reports. Confeund with life etreee. Monroe (1983) identified possible reasons for the interpretive difficulties in the life stress-psychological distress (i.e., neuroticism) relationship. First, people who experience a larger number of psychological symptoms may be more likely to report greater levels of life stress. Second, the item content of measures of life stress and psychological distress are somewhat similar and, thus, may reflect a common nomological network. With regard to the Monroe's first issue, neuroticism has been associated with negative affect. This relationship may be responsible for differences in how different people perceive similar life events. That is, relative to non- neurotic (i.e., stable) persons, neurotic individuals may have a greater tendency to view similar life events in a more negative light. There is both theoretical and empirical evidence to support the notion that emotionally distressed (i.e., neurotic) individuals are more likely to report a higher degree of life stress. From a theoretical perspective, H.J. 33 Eysenck proposed that neurotic persons are more likely to experience more negative affect than stable persons (refer to Eysenck & Eysenck, 1985). Gray's (1981) psychobiological theory was an expansion of Eysenck's theory. He proposed that there are two neurologically-based motivational systems. One is related to reward (i.e., behavioral activation system) and the other is related to punishment (i.e., behavioral inhibition system). Gray hypothesized that neurotics are more sensitive to the inhibition system than are stable persons. This difference in sensitivity to positive and negative life events has been supported by other theorists and researchers (McCrae & Costa, 1991; Strelau, 1987; Tellegan, 1985), but not until recently has this notion been tested empirically (e.g., Larsen & Ketelaar, 1991). From an empirical perspective, Monroe's hypothesis that psychologically distressed people report more life stress is well-documented (see Ormel & Wohlfarth, 1991 for a list of articles published on this topic). In general, correlations between measures of neuroticism and life stress have consistently been reported to range from .40 to .58 in magnitude (Dohrenwend & Shrout, 1985; Kanner, Coyne, Schaeffer, & Lazarus, 1981; Kohn, Lafreniere, & Gurevich, 1991; Watson, 1988). These correlations suggest that neuroticism may influence how life stress is appraised. There is evidence to support this idea. In specific, neurotic persons show 34 greater emotional reactivity to negative situations and less reactivity to positive events (Larsen & Ketelaar, 1991). Neurotic people may perceive the same events as more demanding or threatening (Lazarus & Folkman, 1984) or they may lack the ability to cope with stressful life events (McCrae & Costa, 1986; Tellegan, 1985). They are more sensitive to minor failures, frustration, and daily events than are stable people (Watson & Clark, 1984). Persons scoring high in neuroticism report a larger number of negative events (Aldwin, Levenson, Spiro, & Bosse, 1989; Watson & Clark, 1984), they perceive the events as having a greater impact on their lives (watson, 1988; Watson & Clark, 1984), and they perceive the impact of these events as persisting over a longer period of time than do stable people (Watson & Clark, 1984). Related to Monroe's second issue, the neuroticism confound is clearly related to the content of the items found in life stress questionnaires (Brett, Brief, Burke, George, & Webster, 1990; Dohrenwend et al., 1984; Kohn et al., 1991; Schroeder & Costa, 1984). Item content between measures of major life events and daily hassles with measures of psychological distress are similar. For example, Holmes and Rahe's measure of major life events has been criticized because the majority of its items can be interpreted as symptoms of physical or mental illness (Hudgens, 1974; Schroeder & Costa, 1984). Historically, this confound has led researchers to overestimate the true 35 relationship between life stress and a variety of outcome measures (Schroeder & Costa, 1984; Watson & Pennebaker, 1989). Thus, life stress measures may be more of an indication of psychological functioning than an actual cause of such problems. I Efforts have been made to evaluate and address this potential confound (e.g., Delongis, Coyne, Dakof, Folkman, & Lazarus, 1982; Monroe, 1983; Rowlison & Felner, 1988; Schroeder & Costa, 1984). In one study, Schroeder and Costa (1984) found that confounded life events correlated with health outcomes whereas there was no significant relationship with unconfounded items. While some researchers found supportive evidence for Schroeder and Costa's results (Brett et al., 1990), others were not able to confirm those findings (Maddi, Bartone, & Puccetti, 1987). One of the more heated and interesting debates on this issue involved two research teams who are active in the area of life stress assessment. Even after identifying a significant number of items overlapping between the Hassles Scale and the Hopkins Symptom Checklist (a measure of psychological functioning), Delongis and her colleagues decided against deleting these items (Delongis et al., 1982). Their rationale for not modifying the Hassles Scale was that both versions of the questionnaire correlated .99 with each other. Furthermore, the scale was significantly related to psychological distress whether or not those items 36 were included in the scale. A series of rebuttals followed this publication (e.g., Dohrenwend et al., 1984; Dohrenwend & Shrout, 1985; Lazarus, Delongis, Folkman, & Gruen, 1985). This debate was never clearly resolved, but was beneficial in that it generated a number of conceptual recommendations for measuring life stress (e.g., Dohrenwend, Link, Kern, Shrout, & Markowitz, 1990; Dohrenwend & Shrout, 1985). First, life stress measures should contain both major and minor life stressors. Second, assessments should cover a brief period of time. Third, measures need to differentiate between events and reactions to events (i.e., whether they are viewed as having a negative or a positive impact on the individual). Fourth, predispositions to life stress (e.g., normal personality characteristics, genetic vulnerability, early experiences) should be included in the assessment process. In summary, the extent and the actual effects of the neuroticism confound on measures of life stress remains unclear. Further examination of this relationship is warranted. Confeunu with illness. There is also evidence indicating that neuroticism is confounded with self-reported illness (Costa & McCrae, 1987; Jorgensen & Richards, 1989; McCrae, Bartone, & Costa, 1976). Correlations range from .30 to .50 (see watson, 1988 for a review) and persist across a variety of health problems (Costa & McCrae, 1980; Watson & Pennebaker, 1989). The relationship does not 37 appear to be influenced by the time frame assessed or by the response format used by the questionnaire (Watson & Pennebaker, 1989). Findings from studies on the relationship between neuroticism and objective measures of illness, however, differ substantially from those based on subjective illness reports. Although there is a clear and consistent relationship between neuroticism and self-reported illness, neuroticism is not correlated with actual disease (Costa & McCrae, 1985a; 1987; Stone & Costa, 1990; Watson & Pennebaker, 1989). In addition, self-reported illnesses are related to physicians' evaluations of health, while neuroticism is not related to physicians' ratings. Furthermore, neuroticism is not usually associated with stress-related deaths, such as those resulting from cancer or heart disease (Keehn, Goldberg, & Beebe, 1974; Shekelle, Raynor, Ostfeld, Garron, Bieliauskas, Liu, et al., 1981). These findings suggest that subjective illness reports may be tapping into two sources of variance: one that is related to actual health problems and another that is related to a more subjective or psychological phenomenon (Costa & McCrae, 1987). The difference in the relationship between neuroticism with self-reported illness and actual illness suggests that neuroticism is intertwined with the psychological phenomenon. This linkage between neuroticism and illness highlights the importance of differentiating between the "distress-prone" personality and the "disease- 38 prone" personality (Stone & Costa, 1990). A number of somewhat overlapping personality-disease models have been posed to more clearly understand the relationship between neuroticism and health reports (e.g., Costa & McCrae, 1985a; Holroyde & Coyne, 1987; Peterson & Seligman, 1987; Suls & Rittenhouse, 1990; Watson & Pennebaker, 1989). Watson and Pennebaker (1989) discussed‘ three such models. The psychosomatic hypothesis states that neuroticism causes health problems. This hypothesis is supported by findings that anxiety, depression, and hostility have been linked to both minor (e.g., headaches, acne) and major (e.g., ulcers, coronary heart disease) health problems (Diamond, 1982; Friedman & Booth-Kewley, 1987; Harrell, 1980). A second model is referred to as the disability hypothesis. This model states that health problems cause emotional distress and dissatisfaction. That is, health problems lead to changes in personality, including an increase in neuroticism. Neuroticism is seen as a negative consequence of disease. Watson and Pennebaker do not support either of these models because of the absence of a significant relationship between neuroticism and objective measures of physical health. With this in mind, they pose a third model: the symptom perception hypothesis. This model states that the correlation between neuroticism and self-reported illness is spurious. The relationship simply indicates that neurotics are more vocal and attentive to their physical sensations 39 than are stable persons. Prior research supports this hypothesis (e.g., Costa & McCrae, 1987; McCrae et al., 1976; Tessler & Mechanic, 1978; Watson, 1988). These findings call into question the results of much of the hardiness research. Self-reported illness is the 'most commonly used dependent variable and was the outcome variable used by Kobasa. The use of subjective measures of illness is often justified by their significant correlation with more objective ratings of illness (e.g., Kobasa et al., 1981; LaRue, Bank, Jarvik, & Hetland, 1979; Pennebaker, 1982). Although these relationships are statistically significant, they tend to be low in magnitude, typically ranging from .30 to .40 (Tessler & Mechanic, 1978) to as low as .14 (McCrae et al., 1976). One exception to this is Kobasa's mean correlation of .89 between self-reported illness and medical records (Kobasa et al., 1981). In order to more clearly understand the effects of hardiness on illness, the potential neuroticism confound must be acknowledged. Two options are available for clarifying the relationship between hardiness and illness reports. One option entails statistically controlling for the effects of neuroticism. A second option includes determining whether the strength of the relationship between hardiness and more subjective illness reports is similar to that of hardiness and more objective measures of illness. Recommendations have been made for developing more objective measures of illness (Costa & McCrae, 1987; Stone & Costa, 40 1990; Watson & Pennebaker, 1989). These include biological markers (e.g., immune system functioning), outcome variables (e.g., objective evidence of pathology, disease incidence and mortality), and illness-related behaviors (e.g., number of physician visits, absences from work). Modifications in research strategies are essential if more valid investigations of hardiness are to be completed. Sex Differeneee Because men and women differ both in physiology and in socialization processes, it seems likely that they would differ in how they cope with life stress and in their propensity to report illness (Baum & Grunberg, 1991; Ratliff-Crain & Baum, 1990). There is evidence suggesting that men and women perceive and cope with life stress in different ways. WOmen tend to overestimate the frequency of negative events and are more likely to view events as serious (Kessler, Brown, & Broman, 1981). Women also tend to avoid threatening information or to reinterpret it in a less threatening manner (Stone & Neale, 1984). They tend to be more self-critical, to be less self~rewarding of their accomplishments (Carver & Ganellan, 1983; Gottlieb, 1982), and to use emotion-focused coping rather than problem- focused coping styles (Stone & Neale, 1984). Given the large body of research on sex differences and coping, it seems logical that hardiness may be expressed differently in men and women. With the exception of one study (i.e., Kobasa, 1982a), 41 Kobasa's scale development and research findings on hardiness were conducted exclusively on male samples. Maddi and his colleagues noted that "males are generally less alienated than females", and that further investigation of this issue was warranted (Maddi, Kobasa, & Hoover, 1979, p. 74). Subsequent research has tended to focus on exclusively male samples (e.g., Allred & Smith, 1989; Contrada, 1989; Funk & Houston, 1987; Westman, 1990), exclusively female samples (e.g., Ganellan & Blaney, 1984; Gill & Harris, 1991; Holt et al., 1987; Rhodewalt & Zone, 1989; Schmied & Lawler; 1986) or mixed-sex samples in which the data were not analyzed separately (e.g., Hull et al., 1987; McNeil et al., 1986; Nowack, 1986; Rhodewalt & Agustsdottir, 1984; Wiebe & McCallum, 1986). As a result, there is limited knowledge about the similarities and differences in how hardiness is expressed in men and women. - There is inconclusive evidence from mixed-sex samples that hardiness is exhibited differently in men and women. Some studies identify sex differences with regard to self- reported illness (e.g., Holahan & Moos, 1985; Wiebe et al., 1991) and others do not (e.g., Manning et al., 1988; Roth et al., 1989). Sex differences have been identified regarding the relationship between hardiness and other outcome variables such as attributional style (Hull et al., 1988), psychological symptoms (Holahan & Moos, 1985; Shepperd & Kashani, 1991), the development of hardiness (Hannah & Morrissey, 1987), and physiological indices (Wiebe, 1991). a T‘ 9-- e»! -4 ...A p-q I)“ n -u-. 5" u-‘ a (‘- ”A. “u- 8., n.- :1 42 No sex differences were identified in a study evaluating the relationship between hardiness and a number of mental health-related outcome variables (e.g., depression, anxiety, quality of life; Manning et al., 1988). The hardiness studies which analyzed data separately for both males and females and used self-reported illness as a primary outcome variable are presented below. Roth and his colleagues examined the relationship between hardiness, life events and a number of physical fitness-related variables among college students (Roth et al., 1989). Men and women did not differ in their degree of hardiness, but significant differences were found in the degree of distress experienced, fitness, exercise, and physical illness. In specific, women reported more physical illnesses and negative life experiences and lower levels of exercise activities and physical fitness than men. Shepperd and Kashani (1991) identified sex differences among a sample of adolescents. Although a stress by hardiness interaction was identified for males, no such relationship was found for females. Males who experienced lower levels of stress reported fewer physical symptoms regardless of their level of commitment or control in comparison to high stress males who reported more symptoms when they were low in commitment or control. Wiebe (1991) examined whether hardiness influenced the appraisal of stressful situations using a controlled laboratory task. No sex differences were identified on z— .l ”I 'h. 'l! 2. 43 perceptions of the task as threatening, positive or negative affect, or frustration tolerance. There were sex differences, however, in physiological responses to the threatening task. Hardy men exhibited lower heart rates than non-hardy men when exposed to the stressful situation. No such differences were found for women. The most extensive evaluation of sex differences in how hardiness is expressed was investigated by Wiebe et al., 1991. Two measures of both hardiness and neuroticism were used. This study identified a number of sex differences among a college sample. Males scored lower than females on neuroticism and illness, and higher than females on a measure of challenge. The hardiness composite was a more valid measure for females than for males. Commitment was equally valid for both sexes. Neuroticism was confounded with the control and challenge components to a relatively equal extent for both men and women. Regression analyses of self-reported illness scores were conducted separately for males and females, with some of the analyses controlling for the effects of neuroticism. No sex differences were identified when neuroticism was not used as a covariate. Results from the Personal Views Survey (PVS) indicated that the hardiness composite, commitment, control, and life stress were predictive of illness. No buffering effects were identified. Controlling for neuroticism, however, produced somewhat different results. Data from the PVS indicated that stress 44 continued to be a significant predictor of illness reports, and there continued to be no hardiness buffering effects for both males and females. The hardiness composite, commitment, and control were no longer significant for women whereas they continued to be important predictors of illness for men. A similar pattern of results emerged from the Hardiness Scale data for females. The results for the male sample changed somewhat in that the control and challenge effects were no longer significant after neuroticism was statistically controlled. The challenge component, however, continued to act as a buffer against the effects of stress even after controlling for neuroticism. In summary, only a handful of studies have addressed the issue of possible sex differences in how hardiness is expressed. The limited amount of research coupled with the inconsistency in the findings warrants further exploration of potential sex differences. Criticisme of the Hardinese gonettuct and Herdineee Research Kobasa's theory that hardiness plays an important role in the stress-illness relationship has stimulated much research in the past decade. Over time, however, numerous concerns about the validity of the research findings have been raised. These criticisms can be categorized into measurement issues and the inappropriate use of statistical designs variables. 45 Meaeurement issues. Three measurement-related criticisms have been made against hardiness research. These include the use of negative indicators to measure each component, numerous modifications in the hardiness questionnaires, and the variety of tools available to measure‘hardiness. Kobasa's initial measurement of hardiness (Kobasa, 1979a; 1979b) included both positive (e.g., Adventurousness, Endurance, Leadership Orientation) and negative indicators (e.g., Alienation from Self; Powerlessness). Her six-scale measure of hardiness and the Revised Hardiness Scale, however, only consisted of negatively phrased items. With the exception of the Personal Views Survey, hardiness has continued to be measured solely through the use of negative indicators. That is, hardiness is defined as the absence of alienation, powerlessness, security, and external locus of control. This measurement strategy may be responsible, in part, for the potential confound with neuroticism. There are conceptual and empirical limitations associated with the use of negative indicators (Funk & Houston, 1987). Attempting to measure the presence of characteristics through negative indicators may be erroneous because one cannot be certain that the scales are measuring the true opposite of that particular characteristic. An alternative explanation is that low scores on a particular scale may be indicative of a neutral response rather than the opposite response. 46 The numerous modifications in the measurement of hardiness scales lend themselves to great confusion when attempting to interpret the research findings. Within a six year period of time, Kobasa used four different combinations of scales to measure hardiness. Eighteen scales were initially selected to measure the three components of hardiness (Kobasa 1979a; 1979b). The results from Kobasa's initial research identified six scales which significantly discriminated between hardy and nonhardy male executives. Although some of these scales were used in subsequent research by Kobasa, she also used scales not previously found to differentiate between hardy and nonhardy people (e.g., Kobasa et al., 1981; Kobasa et al., 1982a; Kobasa et al., 1982b; Kobasa et al., 1983; Kobasa & Puccetti, 1983; Kobasa et al., 1985). With the exception of the deletion of the Cognitive-Structure scale (Kobasa et al., 1981; Kobasa et al., 1982a), Kobasa did not provide a rationale for these modifications. Kobasa's research on hardiness has stimulated the development of four variations of her initial 18-scale assessment battery. Each of these questionnaires assesses general hardiness as well as the commitment, control and challenge components. The original measure of hardiness consisted of six of the initial eighteen scales. Kobasa later deleted the Cognitive-Structure scale, which resulted in a five scale measure referred to as the Hardiness Scale. A 36-item Revised Hardiness Scale was later developed as a 47 result of a principal components factor analysis of the original six scales. This questionnaire contains items from the Cognitive-Structure scale. Finally, a 20-item Abridged Hardiness Scale was developed. Nine items from this questionnaire overlap with the Revised Hardiness Scale. A number of other questionnaires have been developed to measure hardiness. The Personal Views Survey contains both positive and negative indicators of hardiness (Hardiness Institute, 1985). The Health-Related Hardiness Scale (Pollock, 1989) was developed to assess people with specific types of health problems. The Abridged Hardiness Scale has been modified to assess an adolescent population (Morrissey & Hannah, 1986). Others have chosen to use a subset of Kobasa's scales in conjunction with other measures hypothesized to assess some aspect of hardiness (e.g., Holt et al., 1987; Nowack, 1986; Zika & Chamberlain, 1987). The availability of such a large variety of hardiness assessment instruments lends itself to at least two research-related problems. First, it calls into question the validity of the research findings. Different measures may not be assessing the same construct or they may be assessing the same construct but in varying degrees. Second, an inability to replicate research findings may be a result of differences in the measures used. r ' t of t i ic l m h . Hardiness research has also been criticized for its frequent use of ANOVA or ANCOVA designs (e.g., Cohen & Edwards, 1989; Funk & 48 Houston, 1987). These designs are less than optimal for at least two reasons. First, the basic underlying assumption of independence is violated because many of the variables in hardiness research are correlated with each other. Second, many of these variables are continuous in nature. In studies employing ANOVA and ANCOVA, however, hardiness scores are typically dichotomized using a median split method. Multiple regression, path analysis, and structural equation modeling are more appropriate statistical techniques for hardiness research. These methodological designs a) measure the effects of each variable while controlling for the effects of others, b) are designed to be used with continuous variables, and c) are therefore more sensitive (i.e., powerful) methods for hypothesis testing (wampold & Freund, 1987). Funk and Houston's (1987) reanalysis of their hardiness data using ANOVA, ANCOVA, and multiple regression techniques highlights this point. The significant hardiness effects found using ANOVA and ANCOVA were not replicated using multiple regression. Summaty There continues to be an interest in the role personality characteristics play in minimizing the effects of life stress. Hardiness is one such personality Characteristic. Kobasa's research provided a basic framework for understanding the relationship between hardiness, life stress, and self-reported illness. Other 49 researchers have continued with this line of research. As a result of these investigations, a number of criticisms and unanswered questions have surfaced. First, should hardiness be conceptualized and studied as a unidimensional or a multidimensional construct? Factor analyses have produced a two-, three-, or four-factor solutions underlying a single general hardiness factor. Most researchers have followed Kobasa's method of using hardiness composite scores to explore the stress-illness relationship. There is, however, evidence that some of the hardiness components may play a more salient role in protecting people against the effects of stress than other components. Studying the relationship of the components on illness reports may prove to be more enlightening that simply using composite scores. Second, to what degree are hardiness, life stress, and illness-related variables confounded with neuroticism? Correlational investigations indicate that these constructs overlap and yet are not redundant. With regard to hardiness and subjective illness reports, statistically controlling for neuroticism tends to minimize or eliminate the effects of hardiness. Third, is hardiness expressed differently in men and women? The development of measures to assess hardiness and the majority of the research has been conducted on males. Recent studies are beginning to highlight similarities and differences in how hardiness is expressed in men and women. 50 Hypothesized Model To address the above questions, the present study developed and tested a model incorporating measures of hardiness, life stress, and neuroticism in predicting self- reported illness and illness behaviors. The basic structural equation model to be tested consisted of Kobasa's original research paradigm with modifications suggested by recent theoretical and empirical criticisms. The basic hardiness model posits that hardiness has a direct effect on illness and an indirect (i.e., buffering) effect on illness through life stress. Two major modifications were made in Kobasa's original model. First, a more objective measure of illness behaviors (i.e., number of days absent from work for health-related problems; number of visits to a physician; number of hospitalizations) was used in conjunction with a standard measure of self-reported illness. Hardiness and life stress were also expected to have an effect on illness behaviors. Second, a measure of neuroticism was added to the model. Neuroticism was expected to be correlated with hardiness, as well as to have an effect on life stress, self-reported illness, and illness behaviors. Confirmatory factor analyses of the Personal Views Survey and the Revised Hardiness Scale were to be completed to determine the number of components underlying the hardiness construct. The results of these analyses would then influence the design of the structural equation models 51 to be tested. Depending on the results of the factor analyses and evaluation of the covariance matrices for men and women, the structural equation model would be tested in a variety of ways. First, one model would explore hardiness effects based on composite scores for the two measures of hardiness, whereas a second model would be based on composite scores. Second, covariance matrices of the male and female data would be examined to identify possible sex differences. The models would then be tested separately for males and females if such differences are identified. Third, the model would be tested using both frequency scores and severity scores derived from the life stress and self-reported illness measures. The models designed to test the modified hardiness theory based on both composite and component scores are depicted in Figure 1 and Figure 2, respectively. S. XI va ;, m 1 y ‘3 Hardiness Y“ ; Illness 1 51 )4! X; Y3. RHS 5”: $3 19 NBC ' 113 1 64 Y4 Neurotic Y3; 7 Illness Beh ‘BEH a, z a. Figure 1. Proposed Structural Model - Composite Scores 53 Ya Hass —-E, 1 Eu Th 6: Life Stress X: PVS 5. 1 Yu ‘1 Commit Y,, )m RHS Y ¢.. Y " PVS Q l X5 Y1 5' $4 6,, Control A“ Y3: RHS := M g Y's va ’ Y2. X5 ' s; s, 4., Chall Y x. l RHS Y ;=: ¢45 “ Y“ NEOA Q - 1 Y”, X: S; Neurotic Au Xe Y2. PERI '- Ea. Au an a“ 711 Illness 113 Illness Beh 1 I: 5"; ",3 l SIRS BEH Figure 2. Proposed Structural Model - Component Scores CHAPTER III Methodology Subjects Three hundred Michigan State University (MSU) employees were randomly selected to participate in this study. A power analysis was completed to determine the appropriate sample size (Cohen, 1988). This calculation (i.e., based on an alpha level of .05, an effect size of .10, and a power value of .79) estimated a sample size of 130 subjects. This number was then increased to 300 subjects to compensate for the possibility of obtaining a moderate return rate. A weighted sampling procedure was used to obtain a representative sample of the overall MSU employee population as of September 1, 1991. One hundred sixty-eight (168) subjects were obtained from the university support staff pool. Occupations represented in this sub-sample include clerical-technical personnel, maintenance and skilled trades laborers, campus police, and operating engineers. One hundred thirty-two (132) subjects were obtained from the faculty/academic staff pool. Occupations represented in the Second sub-sample include professors, coaches, administrators, extension personnel, and library staff. BeCause the subjects were randomly selected from their 54 55 respective populations, the sample, by definition, is representative of MSU employees. Procedures Four versions of the survey were developed. One version was randomly assigned to each subject. The four ‘versions differed only with respect to the ordering of the measures. The purpose of the four versions was to minimize possible fatigue effects. Each survey contained two measures of each of the following variables: hardiness, life stress, and neuroticism. In addition, subjects completed a measure of self-reported illness and a demographic form. The demographic form contained three items designed to assess illness behaviors. Subjects were contacted by mail approximately three weeks after the beginning of the Fall 1991 semester. This initial mailing consisted of a) a letter explaining the purpose of the study and requesting their participation (see Appendix A), b) an informed consent form (see Appendix B), c) one of four versions of the survey, and d) a stamped return envelope. As recommended by Dillman (1988), three follow-up contacts were completed. A postcard was sent to each of the subjects exactly one week after the initial mailing was completed. The purpose of the postcard was thank those persons who had returned their surveys and to serve as a reminder for those who had not yet returned them. The content of this postcard can be found in Appendix C. A 56 second follow-up mailing to the nonrespondents was completed exactly three weeks after the initial mailing. This mailing included a cover letter informing subjects that their survey had not yet been received and reiterated the information found in the original cover letter (see Appendix D). A second copy of the same version of the survey and another stamped return envelope were also included in this mailing. The final follow-up contact to the nonrespondents occurred seven weeks after the initial mailing. This mailing contained another cover letter (see Appendix E), a copy of the survey, and a stamped return envelope. Tables 1 and 2 contain descriptive information on the demographic and employment-related characteristics of the sample. An overall response rate of 70% and 72% was obtained from the support staff and the faculty/academic staff samples, respectively. Of this overall response rate, completed surveys were returned by 63% of the support staff and 61% of the faculty/academic staff. Of the 185 returned surveys, 105 (56.8%) were returned by support staff and 80 (43.2%) by faculty/academic personnel. With regard to sex differences, 94 (51%) were completed by men and 91 (49%) by women. The sample ranged in age from 23 to 66 years with an average age of 44 years. The majority of the sample was white (90%), married or remarried (75%), and affiliated with either a Protestant (48%) or Catholic (20%) faith. With regard to employment-related variables, the sample 57 Table 1. Sample Demographic Information Variable # % Overall Response Rate Support Staff 118/168 70% Faculty & Academic Staff 95/132 72% Survey Completion Rate Support Staff 105/168 63% Faculty & Academic Staff 80/132 61% Sex Males 94 51% Females 91 49% Age 20 - 27 years 11 6% 28 - 35 years 37 20% 36 - 44 years 44 24% 45 - 53 years 52 28% 54 - 62 years 29 15% 63 years or more 12 7% Ethnicity Caucasian 166 90% African American 7 4% Native American 3 2% Asian American 4 2% Hispanic, Mexican American 2 1% Other 3 2% Marital Status ‘Married 126 68% Remarried 13 7% Widowed 5 3% Separated 2 1% Divorced 17 9% Never Married 22 12% Religious Affiliation Protestant 88 48% Catholic 36 20% Jewish 8 4% Latter-Day Saints 2 1% Other 12 7% None 38 21% N te. N = 185. Numbers and percentages do not sum to 185 when all subjects did not respond to a given item. 58 Table 2. Sample Employment-Related Information Variable # % Educational Level High School 10 5% Some College or 29 16% Specialized Training Associate's Degree 11 6% Bachelor's Degree 28 15% Master's Degree 27 15% Doctorate 78 42% Other ' 2 1% Occupational Category Major Professional 86 47% Lesser Professional 22 12% Administrative Personnel 33 18% Semi-Professional 17 9% Clerical or Sales 1 1% Technical 11 6% Skilled Manuals 7 4% Machine Operators & 1 1% Semi-Skilled Unskilled 6 3% Income $00,000 - $ 9,999 1 ‘ 1% $10,000 - $19,999 13 7% $20,000 - $29,999 52 28% $30,000 - $39,999 37 20% $40,000 - $49,999 17 9% $50,000 - $59,999 15 8% $60,000 - $69,000 17 9% $70,000 or greater 30 16% Length of Time at Present Occupation 0 - 5 years 61 33% 6 - 10 years 36 20% 11 - 15 years 26 14% 16 - 20 years 18 10% 21 - 25 years 22 12% 26 - 30 years 9 5% 31 - 35 years 7 4% 36 or more years 6 3% Note. N = 185. Numbers and percentages do not sum to 185 when all subjects did not respond to a given item. 59 tended to be relatively well-educated and employed in more professional roles within the university. The average annual income ranged from $30,000 to $39,999, and the average length of time at their present occupation was twelve years. In rum n The survey included two measures of hardiness (Personal Views Survey; Revised Hardiness Scale) and life stress (Combined Hassles and Uplifts Scale; PERI Life Events Scale), one measure of neuroticism (Neuroticism scale from the NBC Personality Inventory), two measures of illness (Seriousness of Illness Rating Scale; a measure of illness behaviors), and a demographic and background information form. Hardineee. The Personal Views Survey (PVS; Hardiness Institute, 1985) consists of 50 statements which assess the commitment, control, and challenge components of hardiness. Each statement is answered using a 4-point Likert scale (0 = Not at all true; 3 = Completely true). Higher scores indicate a greater degree of each component. In contrast to the Revised Hardiness Scale, the PVS contains both positive and negative indicators of hardiness. Composite scores are calculated by combining the three component scores. The internal consistency for the composite score range from 0.87 (Wiebe et al., 1991) to 0.90 (Hardiness Institute, 1985). The internal consistency reliability for commitment, control, and challenge are .72, .62, and .70, respectively 60 (Wiebe et al., 1991). Similar values are reported by the Hardiness Institute (1985). Test-retest reliabilities of the PVS over time periods of two weeks or more have been reported to be in the .60's (Hardiness Institute, 1984). A copy of the PVS is found in Appendix F. The Revised Hardiness Scale (RHS) is a 36-item measure consisting of negative indicators of hardiness. The RHS was developed from a factor analysis conducted on Kobasa's original six-scale measure of hardiness (reported in Kobasa et al., 1981; Kobasa, 1982b). All items are answered using a 4-point Likert scale (0 = Not at all true; 3 = Completely true). Higher scores indicate lower levels of hardiness. Subjects receive commitment, control, challenge, and composite scores. The internal consistency for the hardiness composite score is .86 (S. Kobasa, personal communication, November, 1990). Average internal consistency reliabilities for commitment, control, and challenge are .73, .72, and .43, respectively (Hull et al., 1987). Test-retest reliabilities over a three week period of time are .74, .79, .78, and .64 for the composite, commitment, control, and challenge components, respectively (Hull et al., 1987). Findings from the RHS duplicate all the major findings obtained using the original six-scale measure of hardiness (S. Kobasa, personal communication, March, 1991). With the exception of the challenge component, convergent validity for the RHS is demonstrated by its correlation with optimism (range = -.41 61 to -.43), with depression (range = .21 to .45), and with emotional distress (range = .26 to .39). All correlations are in the expected direction. A copy of the RHS is found in Appendix G. Streeeful Life Evente. The 53—item Combined Hassles and Uplifts Scale measures the frequency and severity of daily hassles and uplifts (Lazarus & Folkman, 1989). Only the Hassles responses will be used in this study. Daily hassles are defined as "irritating, frustrating, and distressing demands that to some degree characterize everyday transactions with the environment (Kanner et al., 1981; p. 3). Subjects respond to each item regarding the severity of the event in the past six months. Responses are scored using a 4-point Likert scale (0 = None or not applicable; 3 = A great deal). A copy of this measure is found in Appendix H. This scale correlates moderately with its parent instrument, the Hassles Scale [i.e., .45 for frequency of events and .54 for the severity of events (Young, 1987)]. Predictive validity is demonstrated through research indicating that increases in daily hassles precede increases in dysphoric mood (Kanner et al., 1981) and illness symptoms (DeLongis et al., 1982). The PERI Life Events Scale is the second measure of life stress used in this study (Dohrenwend, Krasnoff, Askenasy, & Dohrenwend, 1978). This scale was originally developed as an interview but has also been administered in 62 a questionnaire format by its developers. The PERI contains 101 statements assessing a variety of major life events. These events are organized into the categories of school, work, love and marriage, having children, family, residence, crime and legal matters, finances, social activities, and health.‘ Subjects respond to each item using a 4-point Likert scale indicating the impact a given life event had-on their lives in the past six months (0 = Not at all severe; 3 = Extremely severe). Scores can be used to provide both measures of both the frequency and severity of life stressors. A copy of the PERI is found in Appendix I. Because this measure is typically used in an interview format, limited psychometric data is available on the questionnaire version of the PERI Life Events Scale. The test-retest reliability for this scale across ten one-month time periods is .25 (Raphael, Cloitre, & Dohrenwend, 1991). Neuroticism. The Neuroticism Scale from the NEO Personality Inventory (NEO-PI; Form S) is the measure of neuroticism used in this study (Costa & McCrae, 1985b; Costa & McCrae, 1989). Statements from this 48-item scale were randomly split into two 24-item measures of neuroticism. Items are answered using a 5-point Likert scale (-2 = Strongly disagree; +2 = Strongly agree). Persons scoring high on neuroticism are prone to experience anger, anxiety, disgust, sadness, embarrassment and other negative emotions. High scores are indicative of persons who are experiencing psychological distress, unrealistic ideas, excessive 63 cravings or urges, and maladaptive coping responses. Items which assessed physical symptoms were intentionally excluded from this questionnaire when it was developed. A copy of this scale is found in Appendix J. Internal consistency reliabilities for the neuroticism scale are .91 and .93 for men and women, respectively (Costa & McCrae, 1985b). Test-retest reliabilities are .87 over a six-month time period (Costa & McCrae, 1985b) and .83 over a six-year time period (Costa & McCrae, 1989). The Neuroticism Scale demonstrates good construct validity. It correlates .75 and .84 with the neuroticism scales of the Eysenck Personality Inventory and the Eysenck Personality Questionnaire (McCrae & Costa, 1985). In addition, it correlates -.70 with the Emotional Stability scale of the Guilford Zimmerman Temperament Survey (Costa & McCrae, 1985b). Predictive validity for this scale is demonstrated by its significant relationship with such coping styles as escapist fantasy, self-blame, withdrawal, and passivity (McCrae & Costa, 1986). Physical Illness Measures. A.modified version of the Seriousness of Illness Rating Scale (SIRS; wyler, Masuda, & Holmes, 1968) was used to assess commonly recognized physical and mental symptoms. Subjects responded to items with regard to the illnesses experienced in past six months. Each item is weighted to indicate the threat to life, discomfort, and disruptiveness of each of the illnesses. This revised self-report checklist contains 111 items. 64 Items excluded were those pertaining to psychiatric disorders, infrequent health problems (e.g., depression, shark bite), and gender—specific disorders (e.g., painful menstruation). Four additional items were added to the SIRS: herpes, Alzheimer's disease, cumulative trauma disorders, and HIV infection. The revised checklist is similar to illness questionnaires used by other researchers interested in hardiness (e.g., Rhodewalt & Zone, 1989; Wiebe et al., 1990). With the exception of minor illnesses (e.g., common cold), there is an 89% agreement rate between the SIRS and medical records (Kobasa et al., 1981). A copy of the SIRS is found in Appendix K. As recommended by Watson and Pennebaker (in press), three items were developed to assess Illness Behaviors. These items include the number of a) appointments with health care professionals in the past six months, b) days absent from work due to physical health problems in the past six months, and c) times hospitalized for physical health problems in the past six months. Responses to these three items are summed to yield an overall score of illness- related behaviors. These can be found in Appendix L as items J, K, and L. Damggraphic and Background Infgrmation Form. Subjects were asked to provide the following demographic information: age, sex, ethnicity, marital status, religious preference, highest level of occupation completed, length of time at present occupation, and income. In addition to demographic 65 information, this form contained the measure of illness behaviors mentioned above and an item inquiring about permanent handicaps and disabilities. A copy of this form can be found in Appendix L. Researgh Hypotheses Confirmatory factor analysis. The results of a confirmatory factor analysis (CFA) are expected to identify three components underlying hardiness. The following scales are predicted to load on the stated components: commitment will consist of the Revised Hardiness Scale (RHS) and Personal Views Survey (PVS) Commitment scales, control will consist of the RHS and PVS Control scales, and challenge will consist of the RHS and PVS Challenge scales. Causal model. Two causal models (represented in Figure 1 and Figure 2) will test the relationship between hardiness, life stress, neuroticism, self-reported illness and illness behaviors. These models depict that life stress mediates the relationship between hardiness and illness. That is, life stress serves as a third variable through which hardiness influences physical illness (for reviews on this concept see Baron & Kenny, 1986; James & Brett, 1984). Figure 1 represents the model derived using hardiness composite scores. Figure 2 represents the model derived from Kobasa's three dimensional conceptualization of hardiness. This is also the model hypothesized to be identified from this study's CFA. Each of these models will be tested separately for men and women. The overall fit of 66 these models will be assessed. Since the proposed model is exploratory and the relationships hypothesized are tentative, the models will be revised as necessary. Data Analysis 1. Descriptive statistics (i.e., mean, standard deviation, skewness, and range) will be calculated for each of the measures of the latent variables as well as for the appropriate demographic variables (i.e., age, education, length of time at present occupation, and annual income). Coefficient alpha, a measure of internal consistency reliability, will be computed for the appropriate measures used in this study. Correlation matrices will be computed to examine to relationship between the variables. The first stage of data analysis will consist of a confirmatory factor analysis (CFA) of the hardiness items to determine their underlying factor structure. This analysis will be completed using LISREL 7. The second stage of the data analysis will consist of developing two structural models, one based on hardiness composite scores and one based on component scores). Data from these models will be analyzed using LISREL 7. The design of the structural model depicted in Figure 2 will be determined by the results of the CFA. To account for measurement error, either two measures or a split scale will be used to assess each 67 variable. The exception to this is the measure of self-reported illness and illness behaviors. One measure will be used to assess each of these variables. A chi-square statistic and other tests provided by the LISREL program will be used to assess the goodness-of- fit of the models. Additional tests of fit will also be used since the chi-square statistic is sensitive to the effects of sample size (Fassinger, 1987; Kerwin, Howard, Maxwell, & Borkowski, 1987; Loehlin, 1987; Marsh, Balla, & McDonald, 1988). Structural equation modeling provides an analysis of causal patterns among latent variables represented by multiple measures (Fassinger, 1987). A full structural model consists of two elements: a structural model which delineates the hypothesized causal structure among the latent variables and a measurement model that identifies relationships between measured variables and latent variables (Fassinger, 1987; Francis, 1988; Kerwin et al., 1987). The data are then transformed into correlation or covariance matrices and a series of regression equations. Next, the model is analyzed to examine its fit with the. population. Finally, further modifications and testing of the theoretical model are indicated by the parameter estimates and goodness-of-fit information (Fassinger, 1987). Structural equation modeling offers a number of 68 advantages over either multiple regression or path analysis. First, structural equation model does not assume that observed variables are measured without error. Second, structural equation modeling allows the researcher to examine how closely the overall model fits the data collected. Third, this type of statistical analysis can be used to identify either simultaneous or bidirectional causation. Post hoc analyses will consist of respectively examining the goodness-of—fit of the model for males and females. cl) Ln CHAPTER IV Results Descriptive Statistigs Prior to beginning the analysis, each variable was examined for missing values, skewness, outliers, and accuracy of data entry. Six data entry errors were identified and subsequently corrected. One error resulted from entering incorrect data, two were due to entering out- of—range values, and three resulted from entering the incorrect number of responses for a given questionnaire. The number of missing values for a scale item ranged from 1 to 13. With regard to the demographic information, missing values were also found for religious affiliation (n = 1). occupation (n = 1), and annual income (n = 3). Table 3 contains the full name, abbreviated name, mean, standard deviation, skewness, and range for each of the variables contained in the proposed analyses. The distribution of the majority of the variables was fairly normal. Positively skewed variables included frequency and severity of illness, illness behaviors, and length of time at the present occupation. The skewness of self-reported illness is expected and indicates that the majority of the sample reported fewer and less severe physical illnesses. A 69 7F: 70 Table 3. Descriptive Statistics for All Variables Variable Name Abbreviation M SD SK Range Personal Views Survey (Composite) PVSCOMP 2.31 .26 -1.01 1.20 - 2.78 Personal Views Survey (Commitment) PVSCOMM 2.49 .32 -1.17 1.31 - 3.00 Personal Views Survey (Control) PVSCONT 2.34 .28 -0.98 1.13 - 2.88 Personal Views Survey (Challenge) PVSCHALL 2.09 .29 0.66 1.12 - 2.65 Revised Hardiness Scale (Composite) RHSCOMP 1.71 .19 -O.68 1.03 - 2.13 Revised Hardiness Scale (Commitment) RHSCOMM 2.79 .25 -1.74 1.67 - 3.00 Revised Hardiness Scale (Control) RHSCONT 1.21 .23 ~1.08 0.44 - 1.63 Revised Hardiness Scale (Challenge) RHSCHALL 1.14 .36 .1 l 0.25 - 2.13 Hassles (Frequency) HASS.FQ 28.28 10.49 -.28 0.00 - 52.00 Hassles (Severity) HASS.SV 1.49 .34 .89 1.00 - 3.00 PERI Life Events Scale (Frequency) PERLFQ 8.08 3.84 .69 0.00 - 20.00 PERI Life Events Scale (Severity) PERLSV 1.31 0.68 -.10 0.00 - 2.78 NEO (Total Score) NEOTTL 71.78 22.87 .25 22.00-82.00 NEO (Random Split #1) NEO.A 31.69 11.86 .28 5.00 - 65.00 NEO (Random Split #2) NEO.B 40.09 12.01 .22 13.00 - 68.00 Seriousness of Illness Survey (Frequency) ILLSXFQ 7.45 4.20 .71 0.00 ‘ 20.00 Seriousness of Illness Survey (Severity) ILLSX.SV 214.36 133.05 .73 0.00-608.00 Illness Behaviors ILLBEH 1.80 1.27 1.90 1.00 - 6.00 1. DD '(3 '1' C 71 few subjects reported a large number of illness behaviors. Inspection of these individuals' surveys indicated that their high scores were a result of having more serious llnesses such as cancer. The distribution of Illness Behaviors is similar to that expected in the general population. The skewness associated with the length of time at the present occupation may be a reflection of the higher degree of education completed by the sample and the relatively young age of this population (Mean = 44 years). More highly educated people enter the job market at a later age and may be less likely to have a lengthy employment history with the institution. Negatively skewed variables include RHS Commitment and educational level. The Commitment scores may reflect the value that subjects place on their work. The skewness associated with educational level is expected given that subjects are employed in a university setting in which a greater percentage of jobs require more skills and training. Income level was rectangularly distributed, indicating that relatively equal proportions of employees were earning a broad range of annual incomes. Table 4 and Table 5 contain the zero-order correlation matrices for the variables used in the proposed structural equation models based on frequency and severity scores, respectively. With regard to frequency scores, hardiness composites were negatively correlated with the frequency of hassles, neuroticism, and the frequency of illness reported. 72 Table 4. Correlations Among Measures in Structural Equation Model - Frequency Variable l 2 3 5 6 7 8 9 1. PVS 1.0 Composite 2. RHS .66" 1.0 Composite 3. Hassles -.31” -.I9' 1.0 Frequency 4. PERI -.14 -.16' .28” 1.0 Frequency 5. NBC -.53" -.38” .26” .06 1.0 Total Score 6. NEORandom -.52" -.37” .25” .09 .96” 1.0 Split #1 7. NBORandom -.49“ -.36” .25” .03 .96” .84”_ 1.0 Split#2 8. Frequency of .20" -.20” .22” .28" .27" .28" .24” 1.0 Illness 9. Illness -.01 .01 -.O6 .09 -.01 -.02 -.05 .09 1.0 Behavior Significance levels: ”p < .01; 'p < .05. 73 Table 5. Correlations Among Measures in Structural Equation Model - Severity Variable l 2 3 4 5 6 7 8 9 1. PVS 1.0 Composite 2. RHS .66" 1.0 Composite 3. Hassles -.24” -.26” 1.0 Severity 4. PERI -.20” -.22" .46” 1.0 Severin 5. NBC Total -.53” -.38" .29” .20" 1.0 Score 6. NEORandom ~52” -.37” .30” .19” .96” 1.0 Split #1 7. NEORandom -.49" -.36" .27” .19“ .96" .84” 1.0 Split#2 8. Severity of -.28" -.27" .34” .26" .32” .33" .29" 1.0 Illness 9. Illness .01 .01 .25" .10 -.01 .02 -.05 .24"1.0 Behavior Significance levels: "p < .01; 'p < .05. 74 These relationships were expected. No significant relationship, however, was found between hardiness and the frequency of PERI life events and the frequency of illness behaviors. As expected, daily hassles were positively associated with neuroticism and the frequency of illness. These scores were not related to the frequency of illness behaviors reported. Unexpectedly, the PERI measure of stressful life events was only correlated with the frequency of self-reported illness. Finally, neuroticism was significantly related to the frequency of self-reported illness but not to the frequency of illness behaviors. Self-reported illness and illness behaviors were positively related to each other. With the exception of the PERI measure, the relationships among the variables in this model are within the magnitude and the direction expected given previous research findings. The frequency of illness behaviors, which is intended to be a more objective illness-related measure, was positively related to the frequency of self- reported illness but not to any other variable contained in the structural model. Somewhat different results were identified with the variables contained in the structural equation model based on severity scores. As expected, significant relationships among hardiness, the severity of life stress, neuroticism, and the severity of illness was identified. Hardiness was not related to illness behaviors. With regard to the 75 measures assessing the severity of life stress, hassles was significantly correlated to neuroticism, the severity of self-reported illness, and illness behaviors. With the exception of illness behaviors, the PERI measure of life stress was also related to these variables. Neuroticism was positively correlated with the severity of self-reported illness but not with illness behaviors. As expected, the severity of self-reported illness was positively related to illness behaviors. A larger number of significant relationships was found among the variables contained in the model based on severity scores than in the model based on frequency scores. These relationships were within the expected magnitude and direction. Somewhat different from the variables found in the frequency model, illness behaviors were significantly related to the severity of hassles as well as to the severity of self-reported illness reported. The correlations between the two measures of each of the underlying latent variables are reported in Table 6. All of the correlations are in the expected direction and are significant at the p < .01 level. The highest correlation was between the two scales measuring neuroticism (r = .84). The lowest correlation was between the two measures of challenge (r = .24). The low magnitude of the correlations between the frequency and severity of life stress suggests that these measures are assessing somewhat similar but not identical constructs. This finding is 76 Table 6. Correlations Between Observed Variables Variable Variable PVS Composite PVS Commitment PVS Control PVS Challenge Hassles Frequency Hassles Severity NEO Random Split #1 Frequency of Illness Severity of Illness RHS Composite RHS Commitment RHS Control RHS Challenge PERI Frequency PERI Severity NEO Random Split #2 Illness Behavior Illness Behavior .66" .72" .67" .24” .28” .46” .84” .24" Significance levels: ”p < .01. 77 expected given that the Hassles Scale tends to measure minor daily stressors whereas the PERI Life Events Schedule assesses major life events. Table 7 contains the correlations between the composite scores and the components of the two hardiness measures. With the exception of the challenge component of the Revised Hardiness Scale (RHS), the correlations are significant and in the expected direction. The component scores of the Personal Views Survey (PVS) are more highly correlated with their composite scores (r's = .78 to .90) than are the RHS component scores with their respective composite score (r's = .65 to .70). The correlations among the PVS components range from .49 to .75, with the commitment and control components being the most highly correlated. The commitment and control components of the RHS are also the most highly correlated (r = .57). The relationship between the RHS composite and the challenge component was substantially higher (r = .65) than what has been reported elsewhere (r = .41, .46; Hull et al., 1987). All other relationships among the RHS components did not reach statistically significant levels. The lack of statistical significance in the correlations of the RHS challenge component with the RHS commitment and challenge components is disappointing and yet expected given recent criticisms of the challenge measure. The internal consistency reliabilities (i.e., coefficient alphas) for all but two of the measures used in this study are reported in Table 8. An internal consistency 78 Table 7. Correlation Matrix Among Hardiness Variables Variable 1 2 3 4 5 6 7 8 l. PVS Composite 1.0 2. PVS Commitment .90" 1.0 3. PVS Control .87" .75" 1.0 4. PVS Challenge .78“ .53- .49' 1.0 5. RHS Composite .66" .63“ .50" .54" 1.0 6. RHS Commitment .68- .72- .61- .36" .66- 1.0 7. RHS Control .71- .64- .67“ .50“ .70‘ .57" 1.0 8. RHS Challenge .07 .03 -.11 .24- .65" -.05 .03 1.0 Significance levels: "p < .01; 'p < .05. 79 Table 8. Internal Consistency of Scales Questionnaire # of items Coefficient Alpha Scale Name PVS Composite 50 .84 PVS Commitment 16 .77 PVS Control 17 .62 PVS Challenge 17 .62 RHS Composite 36 .73 RHS Commitment ' 12 .72 RHS Control 16 .74 RHS Challenge 08 .39 Hassles Frequency 53 .92 Hassles Severity 53 .93 PERI Frequency 101(80) .62 PERI Severity 101(80) .59 NBC Total Score 48 . .91 NBC Random Split #1 24 .84 NBC Random Split #2 24 .84 Frequency of Illness 111(81) .72 Nate. The value in parentheses indicates the actual number of items on which coefficient alpha was calculated. Items having a 0 variance (i.e., those items not experienced by any of the subjects) were not included in the analysis. 80 reliability was not calculated for the severity of self- reported illness measure since its items are proportionally weighted and are ordinal in nature. The coefficient alpha also is not reported for illness behaviors since this measure is only a three-item behavioral indicator of illness-related behaviors. With regard to the measures of hardiness, the coefficient alphas were .84 and .73 for the PVS and the RHS composite scores, respectively. The PVS value is similar to that reported in other research (i.e., alpha = .87 to .90). and the RHS value is somewhat lower (alpha = .86). The values reported in this study are within the acceptable range (i.e., alpha = .70 or greater; Nunnally, 1978) and indicate that the items are adequately assessing a common construct. For the hardiness component scores, the coefficient alphas ranged from .62 to .77 for the PVS and from .39 to.73 for the RHS. The PVS values are similar to and slightly higher than those reported by other researchers. The internal consistency reliabilities for the control (alpha = .62) and challenge (alpha = .62) components of the PVS are slightly below an acceptable level. The RHS values are similar to those reported by previous researchers. The internal consistency of the RHS challenge component is clearly below the acceptable standard (i.e., r = .39). This low value may be an indication that the items are not assessing a similar construct or that the scale is too short 81 to adequately assess the construct (Nunnally, 1978). The coefficient alphas for the frequency and severity of daily hassles were well above the acceptable range (alpha = .92, .93, respectively). However, alpha values were somewhat questionable for the PERI frequency and severity measures (frequency alpha = .62; severity alpha = .59). One possible reason for the lower internal consistency values for the PERI is that this scale assesses a somewhat wider range of life events than does the Hassles Scale. The coefficient alphas for the neuroticism measures (i.e., the split half scales) were .84 for each of the halves. This is well above the acceptable range for internal consistency indices. The coefficient alpha was .73 for the frequency of self-reported illness. This value is also within the acceptable range of values. Infarential Statistics onfirmato fa r nal e . Five principal components factor analyses have been completed on the different versions of the Revised Hardiness Scale (RHS) since Kobasa's initial publication in 1979. Although these analyses did not replicate Kobasa's findings identically, the results of all but two of these analyses (i.e., Funk & Houston, 1987; Morrissey & Hannah, 1986) identified a three- factor solution. The consistency of these findings provided the support for conducting a confirmatory factor analysis on the Revised 82 Hardiness Scale. Although researchers have yet to attempt to replicate the three-factor solution of the Personal Views Survey (PVS), it is logical to conduct a confirmatory factor analysis on this questionnaire for two reasons. First, the PVS was developed by Kobasa in response to criticisms voiced about the RHS and, second, the PVS is hypothesized to consist of the same three constructs of commitment, control, and challenge. Confirmatory factor analyses were conducted on the Personal Views Survey (PVS) and the Revised Hardiness Scale (RHS) using LISREL 7 (Joreskog & Sorbom/SPSS Inc, 1989). The PRELIS program (Marija J. Norusis/SPSS Inc., 1989), a preprocessor of LISREL, was used to prepare the data for the confirmatory factor analyses. The assessment of fit between the hypothesized models and the sample data was completed using a number of goodness-of-fit indices. As recommended, (i.e., Byrne, 1989; Joreskog & Sorbom/SPSS Inc., 1989), the fit of each model was evaluated by examining the a) feasibility of the parameter estimates, b) adequacy of the measurement model, c) goodness-of-fit of the overall model, d) subjective goodness-of—fit indices for the overall model, and e) goodness-of-fit of the individual model parameters. Personal Views Survay. A confirmatory factor analysis was completed on the fifty-item Personal Views Survey (PVS). These findings were based on a sample size of 157 subjects. The resulting factor loadings associated with each of the 83 items for commitment, control, and challenge can be found in Table 9. Unmet? values are similar to communalities and represent the amount of variance accounted for by each item of the questionnaire. The first step in establishing the fit of the model was to determine whether the parameter estimates were reasonable. Negative variances, correlations greater than- 1.0, matrices that are not positive definite, large standard errors (which estimate the precision of each item), and highly correlated parameter estimates indicate that the model is wrong or that there are problems with the data. Examination of the PVS data indicated that, overall, all of the LISREL estimates were reasonable. There were no negative variances, no correlations greater than 1.0, and no positive definite matrices. The standard errors for each of the fifty PVS items ranged from .015 to .124. These small values indicate good precision for each of the items. Only one parameter estimate was correlated greater than .30 with another estimate. These findings supported further exploration of the fit of the model. The second step in evaluating the goodness-of-fit was to establish the adequacy of the measurement model. This was done by examining the squared multiple correlation U8)for each observed variable and the coefficient of determination for all of the observed variables simultaneously. These measures show how well the observed variables (i.e., the items) act, both individually and 84 Table 9. Factor Pattern Restdts of the Confimatory Factor Analyses for the Personal Views Savoy Factor Loadms if Commitment Items (Comm) Comm Cont Chall 1 . I often wake up eager to take up my life where it left of .94 ---- ---- .88 the day before. 2. I find it difficult to inagine getting excited about working. .73 ---- ---- .53 3. Most people who work for a living are just manipulated by .88 ---- ---- .77 theb bosses. 4. No matter how hard you work you never really seem to reach .86 ---- ---- .74 you goals. - 5. It doesn't matter if you work hard at you job since ordy the .94 ---- ---- .88 bosses profit. 6. The most exciting thing for me is my own fantasies. .65 ---- ---- .42 7. I realy look forward to my work. .80 ---- ---- .38 8. It's exciting for me to learn somethhg about myself. .22 ---- ---- .05 9. Thinking of youself as a free person just makes you feel .44 ---- ---- .1 9 frustrated and unhappy. 10. I feel no need to try my best at work sirwe it makes no .57 .--. .--. .32 difference anyway. 1 1 . Most of my life gets wasted debug things that don't mean .90 ---- ---- .81 anythhg. 12. Lots of tines I don't realy know my own mind. .66 ---- ---- .44 13. Ordktery work is just too boring to be worth doing. .34 —--- an .12 14. I fitd it hard to believe people who tell me that the work .50 ---- ---- .25 they do is of vdue to society. 15. I think people believe in individuality oriy to knpress others. .55 ...- ---- .30 16. Politicians run ou' lives. .51 .... .-.. .28 Note. Dashes hdicete not applicable. 85 Table 9 lcornt'dl. Factor Load 1_n_’ Control Items (Cont) Comm Cont Chal 1 . Most of the tine my bosses or superiors wl listen to --« .91 ---- .83 what I have to say. 2. Planning ahead can help avoid most futue problems. ---- .06 ---- .00 3. I uuely feel that I can change what might happen tomorrow ---- .24 ---- .06 by what I do today. 4. No matter how hard I try. my efforts wl accomplish nothing. ---- .71 ~--- .50 5. I feel that it's almost inpoeeible to change my spouse's or ..-- .75 ~-« .56 partner's mind about something. 6. When you marry and have chicken you lneve lost you --- .47 ---- .22 freedom of choice. 7. I believe most of what happens in life is jut meant to ---- .56 ---- .31 happen. 8. Most of the tine It jut doesn‘t pay to try Inerd since things an .76 «- .58 never turn out right anyway. 9. When I make plarna. I'm certain I can make them work. --- .07 -..- .00 10. When I am at work performing a difficult task. I know when ---- -.03 ---- .00 I need to ask for help. 1 1 . I find it's uualy very hard to change a friend's thinking ...- .70 --- .49 about something. 12. When I make a mistake, there's very little I can do to make --- .61 .... .37 things right again. 13. One of the best ways to hands most problems is jut not -... .33 --- .1 1 to think about them. 14. I believe that meet addatae are jut born good at sports. ...- .42 ---- .18 15. When other people get anuy at me it's uufly for no good - .67 ---- .45 reason. 16. I feel that if people ty to Inut me there's uuely not much --.. .56 --- .31 tlnat I can do to stop tlnarn. 1 7. When I'm reprinended at work it uualy seems to be -... .45 ..-- .20 uiudfied. m. Dashes new. not applicable. 86 Table 9 (cont'd). Factor Loadmg' h_2 Challenge Items (Chall) Comm Cont Chall 1 . I like a lot of variety in my work. ---- --- .80 .64 2. I feel uncomfortable if I have to make any changes in my ---- ---- .71 .50 everyday echedtle. 3. No matter what you do, the ”tried and true" ways are always ---- ---- .24 .06 the best. ‘ 4. New laws shoddn't be made if tlney Inut a person's income. ---- ---- .25 .06 5. A person whose mind seldom changes can uually be depended ..-. .1 3 .02 on to have reliable judgment. 5. I don't like conversations when otlnars are confused about ---. ~--- .36 .1 3 what tlney mean to say. 7. I won't answer people's questions untl I am very clear as ---- ---- .21 .04 to what they are asking. 8. It doesn't bother me to step aside for a whle from -... --- .47 .22 something 'm involved in. If I'm asked to do somathirng else. 9. I erioy being with people who are predictable. ---- ---- -.25 .06 10. It botlners me when something unexpected interrupts my --- --- .91 .83 daly routine. 1 1 . I respect rues becaue tlney guide me. ---- ---- .24 .06 12. I don't like things to be uncertain or unpredictable. ..-. ---- .77 .59 13. People who do thei beet ehodd get In! financid sunport ---- --—- .31 .10 from society. 14. I have no ue for theories that are not closely tied to facts. ...- .... .21 .04 15. Changes in routine bother me. --- .--. .83 .69 16. Meet days. life jut isn't very exciting for me. --- ~--- .38 .14 1 7. I want to be sue someone wl take care of me when I get ---- ---- .25 .06 old. Note. Dashes indicate not applicable. 87 together, as measurement instruments for the latent variables (i.e., the factors). The values range from O to 1.0 with larger values indicating that the model is a good representation of the data. The R2 indicates the reliability of each observed variable with respect to its underlying latent construct. It indicates the strength of the linear relationship. The fifty PVS R2 values ranged from 0.001 to 0.606. The large number of R2 values that were low in magnitude (i.e., 28 items had R2 values 5 .30) indicate that the model was poorly fitted. The coefficient of determination demonstrates how well the observed variables simultaneously assess the latent variable or factor. The coefficient of determination for the hypothesized model was 0.988, suggesting that the model is fit well. This high value is misleading in that the coefficient of determination is a biased estimator. It is important to compare this index to the other goodness of fit indices. The third step in evaluating the model was to establish the goodness-of-fit for the overall model. This was done by examining the chi-square statistic CKU, goodness-of-fit index, adjusted goodness-of-fit index, and the root mean square residual. These statistics and other supplemental indicators of goodness-of-fit can be found in Table 10. The X? is a likelihood ratio statistic that tests the 88 Table 10. Goodness-of-Fit Indices for Confirmatory Factor Analyses Model x2 df GFI AGFI RMR xz/df TLI Personal Views Survey 2361.30 1175 .620 .587 .214 2.01 .35 Revised Hardiness Scale 1071.59 594 .720 .686 .138 1.80 .47 No; . df= degrees of freedom; GFI = Goodness—of-Fit Index; AGFI = Adjusted Goodness-of-Fit Index; RMR = Root Mean Square Residual; TLI Tucker-Lewis Index. 89 fit between the proposed model and the actual data. Large )8 values indicate that the fit of the model is poor, whereas small values indicate good fit. The degrees of freedom serve as a standard by which to judge whether the X2 is large or small. The inmmsure is sensitive to sample size and departures from multivariate normality in the observed variables. Large sample sizes and departures from normality tend to inflate the x2 statistic. A significant p value indicates that the hypothesized model did not generate the data. The X2 value for the PVS confirmatory factor analysis model was 2361.30 and significant at the p g .0001 level. This indicated that the model was poorly fitted. The goodness-of-fit index (GFI) indicates the amount of variance and covariance jointly explained by the model. The adjusted goodness-of—fit index (AGFI) is a similar indicator except that it adjusts for the number of degrees of freedom in the model. Both indices range in value from 0.0 to 1.0 with larger values indicating a good fit. The CPI and AGFI were 0.620 and 0.587, respectively. These values indicate that the fit of the model was questionable. The root mean square residual (RMR) assesses the average discrepancy between the covariance matrices and the hypothesized values of these matrices. Values range from 0.0 to 1.0 with smaller values indicating a better fitted model. Byrne (1989) recommends values of less than .05, although she states that erroneous models may have values less than .05. She also cautions that the RMR should not be 90 interpreted in isolation of other indicators for this reason. The RMR of the proposed PVS model was .214, indicating a poorly fitted model. Next, the subjective goodness-of—fit indices for the overall model were examined. Because the x2 ratio is influenced by sample size, other goodness-of—fit indices have been proposed (see Marsh, Balla, & McDonald, 1988 for a discussion and recommendations regarding this issue). Two commonly used indices are the Xz/df ratio and the Tucker- Lewis Index (TLI; Tucker & Lewis, 1973). Values less than 2.0 for the Xz/df ratio suggest an adequate model. For the TLI, absolute values of .90 or greater provide support for the fit of the model. The xz/df ratio for the proposed model was 2.01, suggesting that the fit of the model is questionable. The TLI is a more valid indicator of goodness-of-fit because it is not as sensitive to sample size as the Xz/df ratio. The TLI was .35, indicating that the model was poorly fitted. Because the x2, GFI, AGFI, RMR, xz/df, and TLI are measures of overall fit, they do not identify specific parts of the model that may be misspecified. T-values, normalized residuals along with their associated Q-plot, and the modification indices provide more specific information about the fit of the model. T-values consist of the parameter estimates divided by their standard error. They indicate whether or not a 91 parameter is significantly different from zero. Values greater than 2.0 are considered statistically significant (Byrne, 1989). The PVS t-values ranged from 1.52 to 12.82 with four of the items failing to reach significance. Three of these items were-hypothesized to load on the control factor and one on the challenge factor. These weaknesses were corroborated by the factor loadings and hzvalues in Table 9. Using h2 values equal to or less than .25 as a criterion, there are a total of 24 items that only account for minimal variance. Commitment appears to be the strongest factor, whereas challenge appears to be the weakest. The standardized residuals were also examined to identify items that may have been contributing to the lack of fit in the model. This information indicates the discrepancy of fit between the sample and the hypothesized covariance matrices. These values are analogous to z-scores and represent the number of standard deviations the observed residuals are away from the residuals that would be found in a perfectly fitted model. Values greater than 2.00 provide clues as to which items may be misspecified. There were 427 of a possible 1250 PVS items (34%) with values greater than 2.0. The Q-plot, a graphical depiction of the normalized residuals, provided further support for the lack of fit of the model. The modification indices provide a third indication of the goodness-of-fit for the individual model parameters. 92 These values represent the expected drop in the x? if a particular parameter (i.e., the item with the largest modification index) is set free. According to Long (1983), respecification and reestimation of these values should be guided by theory and not simply driven by the modification indices alone. The improvement of the model fit is suggested by the reduction in the X2 statistic. In an effort to improve the fit of the model, the modification indices were used to identify items to be set free in the LISREL program. The results of this specification search are found in Table 11. The hypothesized three-factor model was revised four times. Although each modification resulted in an improvement in the model, the changes were not large enough to support the model. This was most clearly indicated by the change in the TLI. There was only a .08 improvement in the model across the four modifications. The fourth TLI value of .43 (i.e., the value associated with Model 5) was a clear indication of a deficient model. Rayiaad_fla1uiuaaa_§pala. A parallel set of steps was used to evaluate the factor structure of the Revised Hardiness Scale (RHS). This confirmatory factor analysis was completed on a sample size of 152. The resulting factor loadings and the h2 values can be found in Table 12. The first step in evaluating the fit of the RHS model was to screen the output for negative variances, correlations greater than 1.0, matrices that were not 93 Table 11. Respecification Steps in Model-fitting Process for PVS Competing Models for PVS X2 df Ch-X2 Ch-df xz/df TLI 0 Null Mbdel - Personal 3136.68 1225 ---- ---- 2.56 ---- Views Survey 1 Three Factor Model 2361.30 1175 ---- 50 2.01 .35 2 Mbdel with Lambda X 2307.44 1174 53.86 1 1.97 .38 (46,1) free 3 Mbdel with Lambda x 2272.71 1173 34.73 1 1.94 .39‘ (46,1), (1,2) free 4 Mbdel with Lambda x ‘ 2241.67 1172 31.04 1 1.91 .41 (46,1), (1,2), (3,2) free 5 Model with Lambda X 2206.22 1171 35.45 1 1.88 .43 (46,1), (1,2), (3,2), (23,2) free Nppa. df: degrees of freedom; Ch-Xé :- Change in x7 Ch-df = Change in df; TLI = Tucker-Lewis Index. Table 12. Factor Pattern Reedts of the Confimatory Factor Analyses for the Revised Hardiness Scale 9’4 Factor Load _h_’ Commitment Items IComml Comm Cont Chall 1 . Life Is empty and has no meaning in it for me. .89 ..-. --.- .79 2. Most of life is wasted in meaningless activity. .52 .... mo .27 3. I find it hard to believe people who actualy feel dnat the .55 ---- ---- .30 work they perform Is of value to society. 4. No matter how hard I try. my efforts wl accomplish nothing. .82 ---- ---- .67 5. I find it difficndt to inagine enthusiasm concerning work. .43 ~-- --- .18 6. The hunan'e fabled ablity to think is not really such an .55 -« --- .30 advantage. 7. I am ready interested in the poealnlity of expanding my .21 ---- .--. .04 coneciounaes though drugs. 8. The most exciting thing for me is my own fantasies. .56 ---- .--- .31 9. I wonder why I work at al. .61 .--- --- .37 10. The attempt to know youeelf is a waste of effort. .44 ---- --.. .19 11. Ilongforaeinplelifeinwhichbodlyneedearethemoet .67 ...- -- .45 imporuntthingsand decisionedon’thavetobemade. 12. lfyouhavetowork,youmlghtaewalchooeeacareer .35 --- ~— .12 where you deal with matters of life and death. Note. Dashes indicate not applicable. 95 Table 12 Icont'dl. Factor Loadiggg In: Control lterrne (Cont) Comm Cont Cine! 1 . Politicians control ou lives. ---- .80 ---- .64 2. Most of my activities are determined by what society ~--- .42 ---- .18 demands. 3. No matter how hard you work. you never realy seem to ---- .55 -... .30 reach you goals. 4. I uneets me to go into a situation wifinout krnowing what I ---- .34 an .12 can expect from it. ' 5. Those who work for a living are manbrdated by thei ---- .60 ---- .36 bosses. 6. In the long run. people get the respect finey deeerve in ---- .63 ---- .40 finie world. 7. The idea that most teeclnare are unfai to students is ---- .30 ---- .09 norneenee. 8. Capable people who fai to become leaders have not taken ---- .36 ---- .13 advantage of finei opportunities. 9. Becoming a success is a matter of hard work; luck has --- .58 --.. .34 little or nofining to do with it. 10. In my case. getting what I want has little or nofining to do ---- .54 ..-- .29 with luck. 1 1 . Getting people to do fine right fining depends unorn ablity: ---- .57 ...- .32 luck has little to do with it. 12. There is realy no such finirng as “luck.“ --- .46 ..-- .21 1 3. Wlfin enough effort we can wine out politicd oorruntiorn. ---- .30 ---- .09 14. It is inpoeelnle for me to believe finat charnce and luck play --- .63 ~--- .40 an inportant role in my life. 15. What happens to me is my own doing. ---- .68 ---- .46 16. In fine long run, the people are rasponelnle for bad -.-. .35 ---- .12 goverrnmerntonanationalaswelasonalocalbesis. gag. Daelnee indicate not applicable. 96 Table 12 (cont'd). Factor Loadings In_2 Chalenge Items lCheI) Comm Cont Chel 1 . There are no conditiorns finet jutify erndangering fine healfin. ---- ..-- .75 .56 food, arnd shelter of one's famly or of one's self. 2. Pensions large enough to provide for digrnlfied living are fine -... ~--- .46 .21 rightofalwhenageorllneespreventsonefromworking. 3. I very seldom make detaled plarns. ---- ---- .01 .00 4. I tend to start working on a new task wifinout epernding ---- ---- .06 .00 muchtinefininkirng aboutfinebeetwaytoproceed. 5. Before I ask a question, I figue out exactly what I know ---- ---- .07 .00 eieadyandwhetitislneedtofindout. 6. Orne who does one's beet should expect to receive --- ..-- .16 .03 complete economic support from one's society. 7. My work is carefuly plarnrned and organized before it is ---- ~--- .24 .06 begun. 8. I lie to be wifin people who are unpredictable. ---- --—- .03 .00 Note. Dashes indicate not applicable. 97 positive definite, large standard errors, and highly correlated parameter estimates. No such problems wereidentified. The standard errors ranged from .076 to .104, suggesting that there was adequate precision. Only two of the parameter estimates were correlated greater than .30 with other estimates. Overall, these findings suggest that the LISREL estimates were reasonable and that it was appropriate to proceed in examining the fit of the RHS model. Second, the adequacy of the measurement model was explored using the squared multiple correlation CEO for each observed variable and the coefficient of determination for all of the observed variables simultaneously. The RHS I? values ranged from 0.000 to 0.547. Of the 36 RHS items, 24 of them had R2 values less than .30, indicating that the model is poorly fitted. The coefficient of determination for the RHS model was 0.980 indicating that the proposed model is excellent. Because it is a biased estimate, however, this value is compared to other goodness-of-fit indices. The third step in evaluating the model was to establish the goodness-of-fit for the overall model using the X2 statistic, goodness-of-fit index (GOF), adjusted goodness- of-fit index (AGOF), and the root mean square residual (RMR). All four of these indices concur that the model is poorly fitted. The 1:2 statistic was 1071.59 (p < .0001) , the GFI and AGFI were 0.720 and 0.686, respectively, and the 98 RMR was 0.138. These results can be found in Table 10. Fourth, the subjective goodness-of-fit indices for the overall model were examined. The xz/df ratio for the proposed model is 1.80, just falling into the acceptable range. Again, because the Xz/df ratio is biased, the TLI was calculated. The TLI of .47 indicated that the model clearly was poorly fitted. Finally, the individual parameters were evaluated in an attempt to identify specific parts of the model that were misspecified. The RHS t-values ranged from 0.071 to 10.694. All five of the non-significant items identified were hypothesized to load on the challenge factor. The items hypothesized to load on the commitment and control components all were statistically significant and thus appear to be important to the hypothesized model. Additional weaknesses were identified after examining the factor loadings and h2 values (see Table 12) . Eighteen of the 36 items accounted for less than 25% of the variance. Challenge was clearly the weakest factor. With regard to the standardized residuals, only 104 of a possible 648 RHS items (16%) had values greater than 2.0. The Q-plot of the normalized residuals provided further support for the lack of fit of the model. Next, the modification indices were examined. A series of items were freed in an effort to improve the fit of the model. The results of these respecifications can be found in Table 13. The four modifications in the model did not Table 13. Respecification Steps in Mbdel-fitting Process for RHS 99 Competing Models for RHS x2 df Ch-x2 Ch-df xz/df TLI O Null Mbdel - Revised 1577.38 630 ---- ---- 2.50 ---- Hardiness Scale 1 Three Factor Mbdel 1071.59 594 ---- 36 1.80 .47 2 Model with Lambda X 1050.15 593 21.44 1 1.77 .49 (6,1) free 3 Mbdel with Lambda X 1027.46 592 22.69 1 1.74 .51 (6,1), (4,1) free 4 Mbdel with Lambda X 1009.77 591 17.69 1 1.71 .53 (6,1J (4,1), (8,1) free 5 Mbdel with Lambda X 984.04 589 9.85 1 1.68 .54 (6:1) (4:1): (811)! (3,2) free Note. df= degrees of freedom; Ch-JC2 a: Change in x2; Ch-df a Change in df; TLI = Tucker-Lewis Index. 100 improve its fit substantially. This is demonstrated by a small improvement in the TLI (i.e., a .07 increase) across the four revisions in the model. In summary, the initial results of the confirmatory factor analyses for both the Personal Views Survey and the Revised Hardiness Scale did not support a three-factor solution. Additional attempts to respecify the model did 1 not improve either of the models significantly. As a logical next step, principal components analyses were completed in an effort to clarify the actual factor structure suggested by this data. Principal compgnents analyses (PCA). Given the results of the confirmatory factor analyses, principal components analyses were conducted in an effort to understand the underlying component structure of both the Personal Views Survey (PVS) and the Revised Hardiness Scale (RHS). These analyses were conducted using SPSS (SPSS, Inc., 1990). An oblique factor rotation was used given that a) hardiness theory suggests that the components overlap with each other conceptually and b) the hardiness components are correlated with each other. A principal components analysis (PCA) consists of three basic phases. First, relevant statistics and the correlation matrices are examined to determine the viability of conducting a PCA. Second, the number of components underlying the measure are estimated. Third, the components are rotated to aid in their interpretation. 101 Personal Views Survey (PVS). Two criteria were used to determine whether there was preliminary support for conducting a principal components analysis on the PVS. First, the Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy was examined. The KMO compares the observed correlation coefficients to the partial correlation coefficients. Small differences indicate that principal components techniques-should not be used since the correlations between the items cannot be accounted for by other items. KMO values of .60 or greater are required for good principal components analyses (Tabachnik & Fidell, 1983). The KMO value generated on this initial analysis of the PVS was an acceptable .74. The Bartlett Test of Sphericity (BTS) was the second statistic examined. The BTS indicates whether the correlation matrix among the items is an identity matrix. A significant Bartlett statistic indicates that it is appropriate to conduct a principal components analysis. The statistic generated for the PVS was significant at the p < .0001 level and, thus, was supportive of conducting a PCA. The second stage in conducting a PCA is the component extraction process. The three criteria examined in this stage of the analysis are the number of eigenvalues greater than 1.0, the size of these eigenvalues, and the scree plot. The SPSS program identified sixteen components with eigenvalues greater than 1.0. Greater attention was then given to those components that accounted for larger amounts 102 of the variance. These components provide a more accurate representation of the data. When taken as a whole, five components accounted for 36.4% of the variance. A five- component solution was further supported by the scree plot (i.e., plotted eigenvalues). A distinct change in the slope of the plotted values is an indication of the actual number of components present. There appeared to be a distinct change in the slope between the fifth and sixth components. Communalities were then examined in an effort to improve the fit of the five-component model. A communality is the squared multiple correlation coefficient between an item and all other items in the model. Low values (i.e., < .25) indicate that a particular item is not accounting for a significant proportion of variance in the model and, thus, should be eliminated from the analysis. The results of the five-component solution identified nine items with communalities less than .25. These items were deleted from the measure and the analysis was completed again for the model. The resulting components were then rotated using an oblique rotation. The rotation process emphasizes differences in the loadings and aides in the interpretation of the components. The model appeared to be improved with the nine items deleted. The KMO increased to .76 and the Bartlett test statistic remained statistically significant. The amount of variance accounted for by this model was 41.9%, an increase of 5.5%. 103 Four empirical criteria were also supportive of the five-component model. First, there was an overall increase in the magnitude of the communalities. All the values were greater than .25. Second, the residual correlation matrix was examined. In a good principal components analysis, the values of this matrix are small when there is little difference between the original correlation matrix and the correlation matrix generated by the component loadings. Fifty-eight percent of these values were less than .05. Third, the matrix of partial correlation coefficients, referred to as the anti-image correlation (AIC) matrix, was examined. A partial correlation coefficient is the relationship between two items after controlling for the effects of the other items in the measure. Partial correlation coefficients can be thought of as correlations between the unique components. Small coefficients lend support to the results of the principal components analysis. Only 7.9% of these coefficients were greater than .09 in magnitude. Fourth, the values found on the diagonal of the AIC, an indication of sampling adequacy, were examined. Large values lend support to the adequacy of the analysis, whereas small values indicate that certain items are not contributing significantly to the model. The PVS values ranged from .55 to .89 with a mean value of .73, indicating that the data supported a principal components analysis. More important than the empirical support for the five- component structure is whether the components are 104 conceptually sound. Only items with component loadings of .30 or greater were examined (Tabachnik & Fidell, 1983). Table 14 contains the items associated with each component, their respective loadings, and the components they were originally hypothesized to assess according to Kobasa's theory. The first component accounted for 18.1% of the variance and appeared to measure some combination of external locus of control, hopelessness, or helplessness. This component primarily contained items identified by Kobasa as assessing the control and commitment components. The second component accounted for 7.1% of the variance and assessed internal locus of control over upcoming life events. It was also comprised of a combination of commitment and control items. Component 3 measured adherence to authority or security. It accounted for 6.9% of the variance and was comprised primarily of challenge items. The fourth component contained only challenge items and assessed the degree of comfort with a lack of predictability in life. This component accounted for 5.4% of the variance. The fifth and final component assessed a sense of alienation or a "just world" philosophy of life. It accounted for 4.4% of the variance of the five-component model and was comprised primarily of control and commitment items. Revised Hardiness §cele (RH§). A parallel analysis was completed on the Revised Hardiness Scale (RHS). Two statistics were examined to determine the viability of 105 Table 14. Component Loadings for the Personal Views Survey Component 1 Theoretical Item Component Loading Comgonent C1 C2 C3 C4 C5 Cont I feel that it's almost impossible to change my .67 spouse’s or partner's mind about something. (10) Cont I feel that if people try to hurt me, there's usually .62 not much I can do to stop them. (45) Cont When you marry and have children you have lost .61 your freedom of choice. (13) Comm Lots of times I really don’t know my own mind. (39) .58 Cont When I make a mistake, there’s very little I can do .53 to make things right again. (31) Chall Most days, life just isn't very exciting for me. (46) .52 Comm I find it difficult to imagine getting excited about .51 working. (8) Comm No matter how hard you work, you never really seem .50 to reach your goals. (14) Comm Most people who work for a living are just manipulated .47 by their bosses. (1 1) Cont I find it's usually very hard to change a friend's .39 thinking about something. (28) Comm Ordinary work is just too boring to be worth doing. .32 (41) Note. Comm = Commitment; Cont = Control; Chall = Challenge. 106 Table 14 (cont'd). Component 2 Theoretical Item Component Loading Component C1 C2 C3 C4 C5 Comm I really look forward to my work. (23) -.75 Comm I often wake up eager to take up my life where it left -.73 off the day before. (1) Cent I usually feel that I can change what might happen -.63 tomorrow by what I do today. (5) Cont Planning ahead can help avoid most future problems. (4) -.50 Cont No matter how hard I try, my efforts will accomplish -.46 nothing. (7) Cent Most of the time, my bosses or superiors will listen to -.43 what I have to say. (3) Note. Comm = Commitment; Cont = Control; Chall = Challenge. Table 14 (cont'd). 107 Theoretical Component Chall Chall Chall Cont Chall Component 3 Item C 1 I respect rules because they guide me. (33) New laws shouldn't be made if they hurt a person's income. (12) No matter what you do, the "tried and true" ways are always the best. (9) When I am at work performing a difficult task, I know when I need to ask for help. (25) It doesn't bother me to step aside for a while from something I'm involved in if I'm asked to do something else. (24) I believe most of what happens in life is just meant to happen. (16) Politicians run our lives. (50) Component Loading CZ C3 C4 C5 -.57 -.55 -.53 .53 .51 -.43 —.42 Note. Comm = Commitment; Cont = Control; Chall = Challenge. 108 Table 14 (cont'd). Component 4 Theoretical Item Component Loading Commonent C1 C2 C§ C4 C5 Chall Changes in routine bother me. (43) .80 Chall It bothers me when something unemected interrupts .73 my daily routine. (30) Chall | feel uncomfortable if I have to make any changes in .72 my everyday schedule. (6) Chall I don't like things to be uncertain or unpredictable. .65 (36) - Chall It doesn't bother me to step aside for a while from .50 something I'm involved in if I'm asked to do something else. (24) Chall I like a lot of variety in my work. (2) .39 Z C an 0 . Comm = Commitment; Cont = Control; Chall = Challenge. Table 14 (cont'd). 109 Component 5 Theoretical Item Component Loading Component C1 C2 C3 C4 C5 Cont Most of the time it just doesn't pay to try hard, since .61 things never turn out right anyway. (19) Cont When I’m reprimanded at work, it usually seems .58 unjustified. (48) Comm It doesn't matter if you work hard at your job since .54 only the bosses profit. (17) Chall A person whose mind seldom changes can usually be .51 depended on to have reliable judgment. (15) Cent One of the best ways to handle most problems is just .49 not to think about them. (34) Comm I find it hard to believe people who tell me that the .49 work they do is of value to society. (44) Comm I feel no need to try my best at work since it makes .47 no difference anyway. (32) Comm Most of my life gets wasted doing things that don't .45 mean anything. (38) Cont When other people get angry at me, it's usually for .36 no good reason. (42) Comm The most exciting thing for me is my own fantasies. .35 (20) Chall l have no use for theories that are not closely tied to .34 facts. (40) Note. Comm = Commitment; Cont = Control; Chall = Challenge. 110 conducting a principal components analysis (PCA). The Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy was .73 and, thus, was within the acceptable range. The Bartlett Test of Sphericity was significant (p < .001). Similar to the results of the PVS, these two statistics indicate that it was appropriate to proceed with the PCA. During the component extraction phase, the results from the initial PCA identified twelve components with eigenvalues greater that 1.0. A three—component solution was suggested after examining both the amount of variance accounted for by these components and the scree plot. The initial three—component solution accounted for 28.6% of the variance. Next, the communalities for each of the RHS items were examined. Fourteen items with communalities less than .25 were deleted from the measure, and the PCA was completed again. The results from this second analysis were supported by an increase in the KMO (i.e., .78), a significant Bartlett Test of Sphericity, and an increase in the amount of variance accounted for by the model (i.e., 39.4%). Table 15 contains the items associated with each component, their respective loadings, and the components they were originally thought to assess according to Kobasa's theory. Efforts to conceptualize the components, however, indicated that the RHS was actually comprised of two rather than three components. The first two components were clearly identifiable, whereas the third component was not. 111 Table 15. Component Loadings for the Revised Hardiness Scale Component 1 Theoretical Item C1 C2 C3 Component Comm Most of life is wasted in meaningless activity. (5) .64 Comm I find it hard to believe people who actually feel that .63 the work they perform is of value to society. (9) Cent No matter how hard you work, you never really seem .59 to reach your goals. (1 1) Comm No matter how hard I try, my efforts will accomplish .59 nothing. (12) Chall I very seldom make detailed plans. (6) -.59 Comm I long for a simple life in which bodily needs are the .58 most important things and decisions don't have to be made. (24) Chall I tend to start working on a new task without spending -.55 much time thinking about the best way to proceed. (8) Cont Most of my activities are determined by what society .53 demands. (7) Comm The most exciting thing for me is my own fantasies. (18) .48 Cont Those who work for a living are manipulated by their .47 bosses. (21) Comm If you have to work, you might as well choose a .42 career where you deal with matters of life and death. (25) Comm The human's fabled ability to think is really not such .41 an advantage. (14) Cent What happens to me is my own doing. (35). .32 Nets. Comm = Commitment; Cont = Control; Chall = Challenge. 112 Table 15 (cont’d). Component 2 Theoretical Item C1 C2 C3 Component Cont Becoming a success is a matter of hard work; luck has .76 little or nothing to do with it. (29) Cont Getting people to do the right thing depends upon .73 ability; luck has little to do with it. (31) Cent There is really no such thing as "luck.” (32) .70 Cont It is impossible for me to believe that chance and luck .65 play an important role in my life. (34) Cent In my case, getting what I want has little or nothing .54 to do with luck. (30) Cont Capable people who fail to become leaders have not .52 taken advantage of their opportunities. (28) Cont In the long run, people get the respect they deserve .43 in this world. (26) (loge. Comm = Commitment; Cont = Control; Chall = Challenge. 113 Table 15 (cont'd). Component 3 Theoretical Item C1 C2 C3 Component Chall My work is carefully planned and organized before it is .67 begun.(19) Cont Politicians control our lives. (3) .60 Ngte. Comm = Commitment; Cont = Control; Chall = Challenge. 114 The first component was a combination of external locus of control, hopelessness, or helplessness. The second component measured internal locus of control. The third component only consisted of two items, both of which have relatively high component loadings. They did not, however, make sense conceptually. As such, data from this study suggest that a two-component model best describes the underlying structure of the RHS. In addition to the conceptual clarity of the two- component model, other empirical criteria were also supportive. First, two items continued to have communalities less than .25. These items, however, had reasonable component loadings (e.g., .41, .42) indicating that they should not be deleted from the RHS. Second, the residual correlation matrix identified 48% of the values as being less than .05. Closer examination of this matrix suggested that, overall, the coefficients greater than .05 were relatively low in magnitude. Third, 16% of the AIC coefficients were greater than .09. Finally, the coefficients assessing sampling adequacy ranged from .70 to .87 with a mean value of .77. These statistics provide support for the adequacy of the principal components model. On the basis of these results, it can be concluded that hardiness is not a unidimensional construct. Although neither of the principal components analyses directly supported Kobasa's three component theory, there was some degree of conceptual overlap. The PVS assessed external and 115 internal locus of control, challenge, and alienation, and the RHS measured both external and internal locus of control. Stgscturel Equation Models wish Lasent Variables The purpose of testing structural equation models in this study was to critically evaluate Kobasa's hardiness theory in light of recent criticisms. These criticisms include a lack of attention to measurement issues, the use of homogenous samples, a lack of attention to the role neuroticism plays in this research paradigm, and the need for a more objective measure of illness. Hardiness composite models (Figure 1) and hardiness components models (Figure 2) were proposed to be tested using both frequency and severity scores of the life stress and self-reported illness measures. The analysis, conducted using LISREL 7, was based on listwise deletion using a covariance matrix and a maximum likelihood solution. The results of both the confirmatory and the principal components analyses indicated that the three hypothesized hardiness components could not be identified from the data. This finding indicated that it was premature to examine the relationship among each of the hardiness components, life stress, neuroticism, and illness using the proposed structural equation models based on component scores. As such, only the models based on hardiness composite scores were tested. Efforts to analyze the data using the structural 116 equation models with latent variables were unsuccessful. That is, the LISREL 7 program was not able to identify an admissible solution. The error messages indicated that the program was not able to converge to a solution because certain matrices were not positive definite (i.e., PS matrix: the variance-covariance matrix of the errors associated with the latent endogenous variables, the PH matrix: the variance-covariance matrix associated with the latent exogenous variables, and/or the TE matrix: variance- covariance matrix associated with the error for the observed endogenous variables). A non—positive definite matrix indicates that the model is misspecified in some way (Joreskog & Sorbom/SPSS Inc., 1989). The results of this model indicated that it was a poor fit (X2 = 620.55; df = 11; p < .0001). A variety of possible problematic issues may have been contributing to LISREL's inability to converge to an admissible solution. First, non-positive definite error matrices sometimes occur when the model is underidentified. A model is identified when the unknown parameters are mathematical functions of the known parameters and when these functions can be used to obtain unique solutions (Bollen, 1989). When a model is underidentified there are an infinite number of equations that could generate the observed data (Long, 1983). Or, stated differently, there are too many unknowns in the model and not enough information to solve the underlying mathematical equations. 117 Examination of the proposed model indicated that it was underidentified. The model was revised in an attempt to make the matrices either just identified (i.e., to allow for a single unique solution) or overidentified (i.e., to allow for a goodness-of-fit test of the model). Paths were deleted in a manner that was consistent with both hardiness theory and the mathematical criteria necessary to obtain a just identified or an—overidentified model. As a result of addressing these identification issues, Kobasa's basic research paradigm (i.e., the relationship between hardiness and illness and the buffering relationship between hardiness and life stress) was maintained. This model (Figure 3) also failed to pass the admissability test, and its fit was also poor (x2 = 587.44; df = 13; p < .0001). Second, identification problems also arise when the correlations or covariances that link the latent variables are small (Long, 1983). To determine whether or not this might be interfering with the analysis, the standard errors of the estimates were evaluated. Large values indicate that this may be a problem with the model. One method for alleviating this situation is to set the residuals equal to one another. This strategy also did not produce an acceptable model. Third, it was hypothesized that the LISREL program may have reached a local minimum and, thus, was not able to find a solution. A local minimum can be conceptualized as LISREL's unsuccessful attempt to identify a solution when 118 Y1 Y1. Hass — £1 PERI -— E: Q 1 111 Au i. y, 11. £3 Y3 Hardiness 711 Illness 1 w 5:. )ar X: E $2 (a. Yr: X3 a Y 1 92 113 E4 )4 ’ M )4: $3 1‘ 2‘ Figure 3. Revised Structural Model - Composite Scores 119 there are several possible options (i.e., local minima) available (Joreskog & Sorbom/SPSS Inc., 1989). The program gets 'stuck' in one of these possible solutions and is not able to converge to the most appropriate solution. A local minimum sometimes results when variables or measures in a model are scaled very differently from one another or when they have variances that are very different from one another (C. Turner, personal communication, September, 1992). To address this possibility, the LISREL program was configured to enter a correlation matrix (i.e., a matrix of standardized correlation coefficients) rather than a covariance matrix. The solution still was not admissible. Another possible problem with the model was that two of the latent endogenous variables, self-reported illness and illness behaviors, were being measured using a single indicator. The initial version of the model specified the path between the observed and the latent constructs to be set at 1.0, indicating that the questionnaires had no measurement error. As described by Joreskog and Sorbom (1989), the model was respecified to contain an estimate of measurement error. When this did not alleviate the error statement, the model was revised to include two indicators of illness. That is, both self-reported illness and illness behaviors were specified as measuring the single latent construct of illness. Neither of these strategies produced an admissible solution or an adequate model fit. Convergence problems and the presence of negative 120 variances occur when sample sizes or the number of indicators per latent variable are inadequate (Loehlin, 1987; Fassinger, 1987). The present study had an adequate sample size (n = 176), but only had two indicators per variable. A larger number of indicators is preferable. It was at this point in the analysis process that the measurement portion of the model was deleted and a structural model with observed variables (i.e., single rather than multiple indicators of each variable) was conducted. The NEO split scales were combined and the most internally consistent measures of hardiness (i.e., Personal Views Survey) and life stress (i.e., Hassles Scale) were used in this analysis. The structural equation model was revised slightly prior to beginning the analysis. This revision was based upon a reconceptualization of the illness process and the results of the zero-order correlation matrix. Illness behaviors was placed as the final endogenous variable in the model since people are more likely to experience illness prior to missing work, visiting their physician, or being hospitalized due to physical illness. This revised model is depicted in Figure 4. St c u al a ion Mode with bserved V ri ble The structural model analyses were completed using LISREL 7. They were based on a listwise deletion process, correlation matrices, and maximum likelihood solutions. The coefficients presented in the models are standardized 1.21 SI .X _ Eu X I Y“ Hardiness Yu )3 Vs (bu P” Illness Beh Yo. $2. S; X: bkmnxm, it Figure 4. Revised Structural Model - No Measurement Error 122 values. Models based on both frequency and severity scores were explored for the overall sample. A Box's M test was conducted to determine whether the frequency and severity models should be fitted separately for men and women. This analysis compares the variance ovariance matrices for males and females to determine whether or not they are homogenous (Tabachnik & Fidell, 1983). Results from these analyses can be found in Tables 16 and 17. These global results suggest that there were no differences between men and women with regard to the frequency models, but that there were differences in the models based on severity scores. In order to obtain a clearer understanding of these differences, univariate analysis of variance of each variable by sex was completed for the frequency model (df = 1,180) and the severity model (df = 1,179). These results are presented in Tables 18 and 19. With regard to the variables contained in the model based on frequency scores, women reported higher levels of neuroticism (p < .05), more illness (p < .001), and more illness behaviors (p < .05) than men. With regard to the variables contained in the severity model, women reported more neuroticism (p < .05) and illness behaviors (p < .01), as well as more severe life stress (p < .01) and illness (p < .01) than did men. Men and women did not differ in their level of hardiness. The results of the Box's M test and the univariate analysis of variance suggested that there were sex 123 Table 16. Box's M Test for Gender Based on Frequency Scores Box's M F df p X2 df p 17.43 1.13 15,129902 .324 16.91 15 .324 Table 17. Box's M Test for Gender Based on Severity Scores Box's M F df p X2 df p 29.09 1.88 15,128697 .020 28.22 15 .020 124 Table 18. Univariate ANOVA by Gender - Frequency Scores Variable M M F Significance Men Women Level Hardiness 2.29 2.33 1.49 .223 Neuroticism 67.69 75.54 5.56 .019' Life Stress 27.71 29.10 .81 .371 Illness 6.47 8.62 12.53 .001”‘ Symptoms Illness 1.59 1.99 4.50 .035‘ Behaviors b32119- Significance levels: ‘p < .05; ”'p <.001. 125 Table 19. Univariate ANOVA by Gender - Severity Scores Variable M M F Significance Men Women Level Hardiness 2.29 2.33 1.52 .219 Neuroticism 67.78 75.54 5.38 .022’ Life Stress 1.42 1.57 9.89 .002" Illness 186.79 244.88 9.01 .003” Symptoms Illness 1.54 1.99 6.00 .015’ Behaviors HQL§- Significance levels: ‘p < .05; p < .01. 126 differences in the hardiness paradigm. In addition to exploring the hardiness paradigm with the whole sample based on both frequency and severity scores, structural models were explored separately for men and women. The results from these analyses are provided below. Mo e for h wh l sam l d on fre nc r s. The overall and detailed fit information for the Initial and Final models based on frequency scores for the whole sample can be found in Table 20. The Initial model produced a non- significant X2 08 = 1.33, df = 3, p < .722). A non- significant X2 indicates that the proposed model is similar to that which is expected in the population. Since the X2 statistic is influenced by sample size, other test statistics [i.e., adjusted goodness-of-fit (AGOF) index and root mean square residual (RMR)] were also examined to evaluate the fit of the model. The AGOF and the RMR for this model were .985 and .025, respectively. These values were indicative of a good model. However, as mentioned above, these indices are biased and should be interpreted in conjunction with other goodness-of-fit indices. The detailed fit information contained in Table 20 indicates where improvements in the model could be made. The total coefficient of detenmination for the structural equations was .165. This suggests structural weaknesses in the model as a whole. The equations predicting stress, self-reported illness, and illness behaviors had low squared multiple correlations (e.g., .114, .101 and .011, 127 .wa u c .m 302 .759: 05 Co 52323.: .2 poo: o 3.8%.... £022, 305 2c no:.e> vetoes». n . .NI 032 6332.5 30c... n Inna... uoEocoE>w once... u xwud 32:5 83 n m... .— 2oz p .0. n gem... Pow; oofinxm... n Iwmjzij. -- -- v. P. u mg mp p. mNo. mmm. mmm. v we; 3cm Pow. p u Iwmjtxmj. Pom. F u Imm.3_\m._ F _.o. u com... mom: F0—.uxm=_ u ij<>EoI -- -- we 9. u mm mm —. mNo. mam. Nun. 0 mm; 75.5 .6023; .3055 .223—zoom scones—0m cozaaow .anoom seesaw xepc. at a do Nx .2022 cocooEpoS. peyotopcocm 5.3025 cues. “cox -uoéoocoooo .ncozsotoo cocoEELBoo cocoap< c.2222 seesaw .0 2.205000 mmuoom hocosvoum so pwmmm oHdEmm maonz -- ooflomEHOMGH sam Hoooz mo ancessm .om dance 128 respectively). These low values also indicate structural weaknesses in the model. None of the standardized residuals were greater than 2.0, indicating that there were no serious problems with regard to the relationships between the pairs of variables. T-values and modification indices generated by the LISREL 7 program provide additional clues for improving the fit of the model. T~values less than 2.0 identify paths that are not statistically significant and, if deleted, may improve the overall fit of the model (Fassinger, 1987). The paths between hardiness and illness (t = -.366) as well as illness and illness behaviors (t = 1.401) were not significant. Large modification indices indicate possible measurement error in that particular variable (Joreskog & Sorbom/SPSS Inc., 1989). No large modification indices were identified for this model. Based on the above fit information, one modification was made in the model. The path between hardiness and illness was deleted because it was far from being statistically significant. As a result of this respecification, the overall fit of the model improved. The I? continued to be non-significant (X2==1”46, df = 4, p < .883). The AGOF, RMR, total coefficient of determination, and squared multiple correlations remained virtually the same. The path between life stress and illness became statistically significant. Thus, although the overall fit information was 129 supportive of the model, there was still clear indication of structural weaknesses. Despite the suspicion of such weaknesses between the variables, other changes in the model did not seem either statistically or theoretically justified. The Final model is presented in Figure 5 with its associated parameter values. Model for men based on fregsency scores. The overall and detailed fit information for the Initial and Final models based on frequency scores for men can be found in Table 21. The initial model produced a non-significant X2 08 = .76, df = 3, p < .859), indicating that the proposed model was similar to that which is expected in the population. The AGOF index and the RMR (.984 and .020, respectively) were also supportive of the fit of the model. The detailed fit information contained in Table 21 indicates where improvements in the model could be made. The total coefficient of determination for the structural equations was .208, suggesting structural weaknesses in the model. The equations predicting stress, illness, and illness behaviors also had low squared multiple correlations (e.g., .203, .042, .002, respectively). None of the standardized residuals were greater than 2.0. Although there were no large modification indices identified in this model, the t-values for five of the paths clearly indicated deficiencies in the model. The paths between hardiness and illness (t = -.050), neuroticism and life stress (t = 1.084), neuroticism and illness (t = .672), 130 .886 Y1 -.222* m Hardiness * n .164 .146 Y1 Y5 -.526* .104 .243* '900 -989 Xi A ‘Namodc Figure 5. Final Structural Model - Whole Sample - Frequency ,Mege. Significance level: p"< .05. 131 m :V. n Immjzxmga. mow; u Xm...=.m.. cmh. u X04...\0_~0.:02 $0.. . u 8.2.2.52 n 3.. u 1033.203. one... .1. 03:53 «no. u 53.86562 «00. F H m. 03.0001 u . .N 0.02 .0.0.>0£0m 000c... u Immj. .0E03E>m 000:... u xm....= ”000.5 0.... u w.. .P 0.02 mON. mON. ONO. ONO. mam. new. v on. .05“. vmm. mmm. m on. .055 .00:.0>-u .8205 8.69.665. 0000.09.05 ocozoaom .0c0..0_0..00 0.02.22 00.0300 coceaom 5.30.3.5 5.358.800 .0 «00.05000 330.00... 0.0300 200.). «com .62. E a .6 Nx 6662 2000000000 02.5.3. mmuoom hocmsvoum so ommmm so: -- coflomEHOMGH ham Honor 00 snmsssm .Hm wanna 132 life stress and illness (t = 1.366), and illness and illness behaviors (t = .413) were not statistically significant. Based on the fit information, one modification was made in the model. The path between hardiness and illness was deleted from the model. As a result, the overall fit improved slightly. The X2 continued to be non-significant 08 = .76; df = 4; p < .943). The AGOF improved slightly. The RMR, the total coefficient of determination, and the squared multiple correlations remained unchanged. According to the t-values, two paths improved (i.e., neuroticism to illness; life stress to illness), but they did reach statistical significance. Elimination of other non- significant paths produced models that were fit more poorly. Overall, these findings indicate that the model continued to have structural weaknesses. In summary, although the overall fit is supportive of the model, there were still clear indications of serious structural weaknesses in the model. Four of the paths in the model did not reach statistical significance. Despite these concerns, additional changes in the model did not seem justified. The Final model is presented in Figure 6 with its associated parameter values. W. The overall and detailed fit information for the Initial, Revised, and Final models based on frequency scores for women can be found in Table 22. The Initial model produced a non- significant X2 08 = 3.93, df = 3, p < .270). The AGOF 133 .797 X: Hardiness .122 * -.553 .087 INamodc Figure 6. Final Structural Megs. Significance level: X .160 Y1 y3 .043 .958 .998 Model - Men - Frequency Scores p“ < .05. 13‘1 .00 u c .n 0.02 4000:. 05 .0 00.00.3005 .0. 000: 0 0.00.0... 00.55 0005 0.0 003.0) 08.000: n . .N 0.02 0.030500 002.... N 10044. .0EouaE>0 000:... u X044. 000:0 0..4 u 0.. .— 0.02 0 P0. .. .1. X044=04 005. p- u 10044.3..0502 000. P u X044.\0.uo.302 .00. 1. 10044. 00h. F. 00 F. u X044. H X044_\>0.0I -- : 0.5. u 04 OON. mNO. ohm. .05. m 00. F .05.. «mm. P u Imm4_.\X044. 031—" X044..04 005. p- M 10044.. 0..0.302 000. w u X044.\0..o.302 00¢. P H m.<0_uo:402 3;. T :«O. N 10044. H X044_\>0.0I :. P. u X044. NEE-" 0430.0... -- -- .00. u 04 "ON. 0N0. 050. 000. N 00. 00030”. 000. H 100...... X044. 0 TV. .. u X044<04 000. p u X044..0..o.302 000.. P n 04.0.5502 0.2.. P- hOO. fl 1004.... n X044.\>0.0I 00.N z. ... .1: X044. N3: u 0430.0... u P00.50.00.302 1 P00. .11. 04 05 p. 00C. 0 — 0. CNN. 0 00.0 .00.... .0020?“ .0020... 0.030.000 000003.00 00.00300 330.00: 0.0300 X00... z“. a .0 NX .0003. co..0o...00$. 03.000030 3.30330 0005. .00". -.0-000:0000 51.9.4.3 8.3451503 083.3 0.00.3.2 00.0300 .0 3.0.0...000 mmnoom xodmsvmum no 00000 dmfioz -- c0040§MOMGH sam H0002 mo wudafidm .Nm manmh 135 (.914) and the RMR (.054) were also supportive of the fit of the model. The detailed fit information contained in Table 22 indicates where some improvements in the model could be made, however. The total coefficient of determination for the structural equations was .173, suggesting serious structural weaknesses in the model. The equations predicting stress, illness, and illness behaviors had low squared multiple correlations (e.g., .051, .171, and .007, respectively), another indication of structural weaknesses. None of the standardized residuals were greater than 2.0, indicating that the relationships between the pairs of variables were being fit well. The modification index suggested that the fit of the model might be improved if a path between neuroticism and illness behaviors was added. Based upon this information, a path was added to the model connecting neuroticism to illness behaviors. It is possible that there is a relationship between neuroticism and illness behaviors. Adding this additional path improved the fit of the Revised model (X2== .88; df = 2; p = .643). The AGOF increased to .970 and the RMR decreased to .023. The overall coefficient of determination improved slightly as did the squared multiple correlation for illness behaviors. The magnitude of these values, however, continued to indicate structural weaknesses in the model. The new path approached statistical significance, and the path between 136 :Lllness and illness behaviors improved somewhat. . Aside from the improvement in the model, none of the Eniths in the model were statistically significant. In the ZFinal model, the path between hardiness and life stress was (deleted model because it was particularly weak. This revision resulted in some improvement in the model. The Inodel remained non-significant, as is desirable (X2==1.08; df = 3; p < .781). The total coefficient of determination for the structural equations and the squared multiple correlations remained virtually the same. The importance of the path between neuroticism and life stress improved greatly, reaching statistical significance. The other paths did not reach statistical significance. The modification index did not indicate that further revisions would be helpful. Again, although the overall fit information was supportive of the model, there were still indications of serious structural weaknesses. The Final model is presented in Figure 7 with its associated parameter values. In summary, the results of the models based on frequency scores indicated that the hardiness research paradigm was expressed differently in men and women. Confidence intervals were calculated to determine whether the strength of the common paths in these models were statistically different from each other. Results of these analyses indicated that there were no differences in the strength of the relationship among the common paths. 137 . 951 71 XI Hardiness . 221* ’2. Y; -.561* .154 A .199T . 831 . 959 XI. Neurotic Figure 7. Final Structural Model - Women - Frequency Scores Mote. Significance level: p' < .05. 138 Model for the whole semple besed on severigy seores. A similar model based on the severity of life stress and illness was tested. The overall and detailed fit information for the Initial and Final models can be found in Table 23. The Initial model produced a significant X2 0? = 9.86, df = 3, p < .020), indicating that the model is poorly fitted. The AGOF and the RMR were .896 and .052, respectively, and thus were supportive of the model. The detailed fit information contained in Table 23 indicates where improvements in the model could be made. The total coefficient of determination for the structural equations was .175, suggesting serious structural weaknesses in the model. The equations predicting life stress, illness, and illness behaviors also had low squared multiple correlations (e.g., .100, .192 and .056, respectively). This was another indication of structural weaknesses. In addition, the standardized residual between life stress and illness behavior was greater than 2.0, indicating thatthe relationship between these variables were not being fit well. The t-values associated with this model indicate that the paths between hardiness and life stress (t = -1.386) and hardiness and illness (t = -1.422) were not significant. There was also a high modification index for one indicator, the relationship between life stress and illness behaviors. Based upon this fit information, the model was modified by adding a path connecting illness behaviors to life stress. It seemed possible that experiencing more sick 139 .90.. u 0 .m 0.02 40000. 00. .0 000000.000. .0. 0000 0 0.00.00. 00.02. 000... 0.0 003.0> 08.0000 u . .N 0.02 0.030000 0000... n 20044. u0.00.0005 0000... u X04... "002.0 0..4 u 04 .. 0.02 «be. p- n 0044.30.01 000. n 1004... 00¢. .. 00 F. n X044. u 0430.0: -- -- 00 p. u 0.. OON. «.NO. 500. 00... N 00. p .00.“. NNO. T u X03306: 000. u 100...: 000..- 5.0 0000 N0..HX044. u 0430.0... u 04.10044. u 10044.04 cop. u 04 ms F. N00. 000. ONO. m 00.0 .0...0. .003.0>-. .000.00. .0.030.000 00000300 00.0300 .030.000 0.0300 X000. 03.. a .0 «x .0005. 000000.00.)— 00~.0.0000.0 .0.3.03..0 000—2 .000 -.0-00000000 .0I0Io..0..0|..00 00..00.E.0._0.O. 00.03.04. 0.0.03.2 00.0300 .0 .00.0...000 mmuoom huflum>wm no 00000 magfimm waosz -- coflumenowcH 0.0 H0002 no 0006830 .mN 0.509 140 days, visits to physicians, and hospitalizations would influence the severity of life stress experienced. As a result of this modification, the fit of the Final Model improved greatly. The X2 became non-significant (X2 = 1.40, df = 2; p < .497). -The AGOF increased to .977, and the RMR decreased to .027. In general, there were improvements in the total coefficient of determination and the squared multiple correlations-associated with life stress. The magnitude of these values, however, continued to indicate the presence of structural weaknesses in the model. The standardized residuals were all less than 2.0, indicating that the relationships between the variables were now being fit well in this model. Finally, the t-values improved, but they did not reach statistical significance. Eliminating these paths negatively influenced the fit of the model. For this reason, these paths were not deleted. Although the overall fit information is supportive of this model, there is still some indication of structural weaknesses. Despite these suspicions, additional changes in the model did not seem statistically or theoretically warranted. The final model is presented in Figure 8 with its associated parameter estimates. Mod 1 or n on ve ' . The overall and detailed fit information for the Initial, Revised, and Final models based on severity scores for men can be found in Table 24. The Initial model produced a non-significant X2 0? = 2.87, df = 3, p < .412), indicating that the model 141 . 210 * . 831 M -.116T XI Hardiness . 240* Y2. Y3 . Illness Beh -.526* -192 * . 214 . 807 . 9 7 X2 ‘Nmmnfic Figure 8. Final Structural Model - Whole Sample - Severity Me. Significance level: p' < .05. 142 .Nm " C .M Ouoz 40000. 00. .0 02.00.0000,. .0. 0000 0 0.00.00. 00.0.5 0000. 0.0 003_0> 03.0000 n . .N 0.02 .0.0.>0000 0000... u 10044. 8.00.0050 0000... u X044. 802.0 8.4 u 04 .— 0.02 :0.— " 000.....x0j. 000. u :00... 000.. 03.38.... u 103...... -- -- 000. u 0.. 0.... m .0. 000. 000. v ... .9... 00.. . u $04....x0...= ~00. . u 1004....04 3.. . n xw...=2.982 000.- u 002.932 0.0. u 100...: 9.0.- .0 .. u X04... u 0.0.4.320: -- -- 00 .. u 0.. ~00. 0N0. .00. .00. N 00.. 02.3.... X... . u x04.._3..osoz 000.- u 04.26.82 000. u 000.... 03..- 00.0 .0 .. u x0... u 0.0.4.305: n 00:00.... -- 00 .. u 0. $0. 000. 08. 0 S. 0 .0.~ 0...... 0003—6); 000055 agozgmom UCOBODUM cotanvw inguom 0.030w XOUC. Zn. n— :0 «X Etc.)— COZOOEUO—Z fiDfi—EOUCOHW 3.3.0:.um 200$— uoom -.0-000cn.000 .333. a 088.3 0.0.03.2 00.0300 .0 .00.0...0oo mmuoom xuflum>mm no 00000 cm: -- 00.008H00GH 0.0 H0002 mo >H08850 .00 04209 143 was fitting the data well. The AGOF (.939) and the RMR (.035) were also supportive of the fit. The detailed fit information contained in Table 24 indicates where improvements in the model could be made. The total coefficient of determination for the structural equations was .232. This suggests structural weaknesses in the model. The equations predicting life stress, illness, and illness behaviors had low squared multiple correlations (e.g., .186, .147, .056, respectively), another indication of structural weaknesses. None of the standardized residuals were significantly greater than 2.0, indicating that the relationships between the pairs of variables were being fit well. The Revised model was based on the modification index; it consisted of adding a path connecting illness behaviors to life stress. It made conceptual sense that increased illness behaviors could lead to higher levels of perceived life stress. As a result of this modification in the model, the overall fit of the model improved somewhat (X2==1002; df = 2; p < .601). The fit indices remained virtually the same and continued to indicate the presence of structural weaknesses in the model. The Final model was respecified by deleting the paths between hardiness and illness and between neuroticism and life stress. These two paths were clearly not approaching statistical significance in the model. This model appeared to be a slight improvement over the previous one (X’==.71; 144 df = 4; p < .950). Global and specific fit indices improved somewhat. In specific, the squared multiple correlation associated with life stress improved. The paths between illness behaviors and life stress and between illness and illness behaviors also improved slightly but still failed to reach statistical significance. Thus, although the overall fit information is supportive of the model, there are still indications of structural weaknesses in the model. Despite these concerns, other changes in the model did not seem to be statistically or theoretically warranted. The Final model is presented in Figure 9 with its associated parameter values. Model for women based on severity scores. The overall and detailed fit information for the Initial, Revised, and Final models based on severity scores for women can be found in Table 25. The Initial model produced a significant X2 D? = 10.15, df = 3, p < .017), indicating that the model was not fit well. The AGOF (.791) and the RMR (.071) also suggested that the model could be improved upon. The detailed fit information contained in Table 25 indicates where some improvements in the model could be made. The total coefficient of determination for the structural equations was .244, suggesting structural weaknesses in the model. The equations predicting stress, illness, and illness behaviors had low squared multiple correlations (e.g., .123, .224 and .033, respectively), another indication of structural weaknesses. In addition, 145 .780 X: Hardiness -.552 Xi Neurone Life Stress y. .152T Figure 9. Final Structural Model - Men - Severity Scores EQEQ- Significance level: p' < .05. % 1 .mm H c .n 802 ._ovoE 05 .0 cos-8:59: .3 too: a 2855 £023 305 2c 82.; potency. u . .N. 202 6332.3 noon... n Immj. umEoEE>w «mac... n xwfi. 63.5 23 n ma .— 202 wt... .I. Iwmjtij. «no. u Iwmj. 0mm; oFN.HXm...__ u mains: -- -- hm p. n m.— m 5. 0N0. mom. ecu. N 5.— Eat Now; u Immj