I' I IIII d—.- _- ::--- “L_ -2 1:— ‘ «f _——' ;—.:_-; ‘_‘-.___:_. TTIII IF '.II IIIII‘ INN I I III . I II I III II I: .7 III ' III II I. I “IIIIIII IIIIIIIIII III. II ., I. III III II I II I M I w: III I III III I IIIIIII IIIIII' .»'II.. I I. I .IIIII III. . . 1"th i ' I II I IIIIIIII III» II.IIII II'I III I I w W": I II III II ”II. III .I III II II I II \I III I III. IIIIIII 1"““‘.".'i"‘ II . 5155:4122: ‘ :_ . . IIII'I. I" I ' I” 'I E- SIII‘: “IIIIIITI , '.I. ,.I. I III: 4— “H: JW’ .1ch III II "—f- III II II IIIII III III... III III ‘.I'II‘ ’IIII' .. I. .JIIII. 'YI. :_. .II . . .mI ;' .. 532-88., ‘1‘ ‘.III. II II IIIHI IIII'I. . IIII III»'» .I‘III.II.;I. II '37 "i I.II‘ I I I‘III'I.‘ III III II IIIII'I.';IIIIIIIIII'II.'.33I:IziI‘ ~.'» III“ III I »» I...» ;;».II,...I.»IIIIsa-I: I "I I . ... “I!“ II‘III I».~:IIII I». I II... IIIII...“ .II. . .I... II“ I .1... III III-III III; I III‘ I.1 I III III II| I I II .I .Il....1‘II. .. IIIIIIIIIIIIIHI ..|.IIII I I “WWII“ II I” II I.. . .I.. ‘.»."I » . I ”I. . I I".IIII‘.II..I..,. I I. .I . 33-, ,_ I ' I.»"II ~ «It. I» I. :..»...‘.IIIIiII-I I v H II III‘II'. II IIIIIII ‘ IIIII'III»I II III: II III’IIII III-“II. l:"I .» . :I _ III.“ I III , I...“ I ”H. ' ;|I III. N ‘II.II;‘.':II. I135” 1‘: . .I III. IIIIII I ... ‘.I-I:III.I I 'III ‘.I ;.',.:II..’ III. II‘I III II I TI . IIH‘IIIIIII IIIIII I.. 'IIIIIIII IIIIII III: III" I. I III 'IIi..II IIII I W IIIJ' III II IIIIII II. I IIII' III-J! I: I..‘I""III:»-I-.. 'I'I. ..» ‘;‘II.I. II .' II III.» 'IIIII» IIII’I'IIIII."IIII.- t .. 4. . '. .‘I;"WIIIIIII».'II» II. II I. ‘.I‘II‘I'. I II' I" I "III..I.III....I..I. IIIIIII I I I..IIIIIII‘I'.II IIIIII II II II.III‘I..‘. .II' IIIIII . IIIIII h... .IlIIIIII IIII'I IIII‘. IIII I. .II'I' I I ”III" III“ 'I . I ......I II. I»II"I."' I II 'I'I‘IIIIIIIII'I "I 1» ‘.IIII' I IIIII IIII IIIIII II. .III III ». .»-. » I » .. I ‘.II I» III IIILI I. ‘ 'I‘ “I .. ..:I.II‘ I . .IIIIIIIIII “II | .IIII '. I h .. I H ' I .l ‘IIIII .. IIIII. IIIIIII. l'. {VIII} {In “IF I. I II. » ' I. . .I»..».. IIIII-II I"" 'I IIIIII. . III . I I. ‘ I... I »IIIIII. II..I,;..II III III III! IIIIIIIIII II. .III b. .'I I l I ' .... - ». .. »III ...I III'IIII THE?”— lllfljlfllfil{HM/IIHIHIIH!MINIMUM!)Ill/1111M 1/ 2 10473 8186 This is to certify that the thesis entitled An Appraisal of Nonsurvey and Minimum Survey Techniques For Estimating the Lansing Tri-County Region Input-Output Tabies presented by Lutfi Ibrahim Nasoetion has been accepted towards fulfillment of the requirements for Doctor of Phi]osoohv degmmin Resource Deveiopment W/M Major profeén Date_J_ul)L25_._L9_Z9__ 0-7639 .i ”ha-1 LIBRAF. Y a ,AiChlgafi S 51h: 1: ‘.Inivcrsiiy {~49 . ‘ / m v.29“; 2.3 ”EM—'— OVERDUE FINES: 25¢ per day per item RETURNING LIBRARY MATERIALS: Place in book return to remove charge from circulation record ”W W %m; 5 e113 mzax93‘33 . iv 52% ’ 300 ‘7‘! “fig"? 43:22: '- f2: ‘ 3L? i 7 “w _W '_ ' Tom'ibifiooz, 1.4.]! 118%”. s ‘ ’WCWW ’ . 322 ’3 AN APPRAISAL OF NONSURVEY AND MINIMUM SURVEY TECHNIQUES FOR ESTIMATING THE LANSING TRI-COUNTY REGION INPUT-OUTPUT TABLES BY Lutfi Ibrahim Nasoetion A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Resource Development 1979 ABSTRACT AN APPRAISAL OF NONSURVEY AND MINIMUM SURVEY TECHNIQUES FOR ESTIMATING THE LANSING TRI-COUNTY REGION INPUT-OUTPUT TABLES BY Lutfi Ibrahim Nasoetion The application of a regional input-output model is invariably a compromise between the desire for theoretical satisfaction and comprehensive data coverage, on the one hand, and the need to economise on time and resources on the other. Nonsurvey and minimum survey techniques for constructing regional input-output tables are attractive to regional analysis because of the relatively smaller cost involved as compared with direct survey techniques. Yet, attempts at such nonsurvey and minimum survey tech- niques have not been successful. The distance between the survey and the nonsurvey direct cofficients matrices is quite large in absolute terms. In fact, until recently there were no accurate comparisons between tables con— structed by survey and nonsurvey methods. Thus, a study on nonsurvey and minimum survey techniques of construct- ing regional input-output tables would likely contribute greatly to progress in regional economics. In order to make this contribution, the relative efficiency of six reducing techniques--(l) Simple Loca- tion Quotient, (2) Purchase Only Location Quotient, (3) Cross Industry Quotient, (4) Modified Supply-Demand Lutfi Ibrahim Nasoetion Pool, (5) RAS, and (6) Schaffer—Chu Iterative Procedure-- in estimating the Tri-County Region's (Eaton, Clinton and Ingham Counties of Michigan) direct coefficients, the Type I and Type II income multipliers were compared and eval- uated. The relative efficiency of each reducing technique was measured as distance or relative distance between the estimated direct coefficients, the Type I and the Type II income multipliers, and the "True" direct coefficients, the Type I and the Type II income multipliers computed from the survey table. In judging which of the estimated coefficients and multipliers are closest to the equival- ent survey based table, three comparative tests were used: (1) The mean absolute percentage deviation, (2) The mean of relative change, and (3) The mean of similarity index. In estimating the regional direct coefficients, none of the reducing techniques appraised in this study were entirely satisfactory. The distance between the survey and nonsurvey matrices was too large, to be accptable to the author. It was apparent that the RAS technique pro- duced superior estimates when judged by the three compar- ative techniques. This is not entirely surprising, given the technique employs a certain amount of survey data. Among the purely nonsurvey techniques, the Schaffer-Chu Iterative procedure and the Cross Industry Quotient tech- nique emerged as being superior to the other nonsurvey Lutfi Ibrahim Nasoetion techniques, when judged by the mean of similarity index, which is the most ideal comparison test employed in this study. This study revealed that the relative efficiency of the six reducing techniques was better when they were used to estimate the Type I and the Type II multipliers, than when they were used to estimate the regional direct coefficient. It was observed, that the RAS technique pro- duced the best estimates of the Type I and the Type II multipliers when judged by any comparison tests employed in this study. Among the purely nonsurvey techniques, the Schaffer-Chu iterative procedure and the Cross Industry Quotient produced the best simulation of the Type I and Type II income multipliers. In estimating regional direct coefficients, Type I, and Type II income multipliers, it was found that the greatest deviations occurred in three groups of sectors: (1) Sectors in which the regional economy is highly specialized, (2) Sectors of primary activities, and (3) Sectors which have been obtained through a high de- gree of aggregation. In order to obtain more acceptable results, it was recommended to use field surveys to ob— tain the regional direct coefficients of these three groups of sectors. Dedicated to Ros and Rina Nasoetion ii ACKNOWLEDGMENTS The writer wishes to express his sincere apprecia- tion to Dr. Milton H. Steinmueller who, as chairman of the Guidance Committee, gave his help, encouragement, support and friendship throughout my entire doctoral pro- gram. Gratitude is also expressed to Dr. Daniel E. Chap- pelle for guiding the thesis work, and to Dr. Raleigh Barlowe and Dr. Garland P. Wood for serving on the Guid- ance Committee and reading the thesis. In addition, the writer is indebted to first, Bogar Agricultural University which sent him to Michigan State University to pursue the program-~special thanks here to Dr. Oetik Koswara, and Dr. Andi Glakim Nasoetion, Presi- dent of Bogar Agricultural University, who continuously encouraged the completion of the study; and second to MUCIA_AID mission to Indonesia which provided the funds that enabled the writer to stay in the United States dur- ing his masters and doctoral program. The writer wishes to acknowledge his friends in his home country and the United States--special thanks to Dedi Fardiaz, Uben Parhusip, Hadi K. Purwadaria, and Lukito Sukahar, who have made innumerable contributions to the enjoyment and accomplishment of this endeavor. Finally, to my wife Ros Nasoetion and my child Karina Nasoetion, who gave and still will give unenduring love, encouragement, and understanding, the writer offers his warmest gratitude. iii LIST OF CHAPTER I. II. III. TABLE OF CONTENTS TABLES O O O O O O O O O O O O O O O O O O 0 INTRODUCTION . . . . . . . . . . . . . . . . Problem Setting . . . . . . . . . . . Objectives of the Study . . . . . . . . Limitations of the Study . . . . . . . . Outline . . . . . . . . . . . . . . . . LITERATURE REVIEW . . . . . . . . . . . . . The Theoretical Foundations of Input-Output AnaIYSiS O O O O O I I O O O I O O O O I O Transactions Matrix . . . . . . . . . . Input Coefficients . . . . . . . . . . . Gross Output Needed to Produce Given Final Demand . . . . . . . . . . . . . Prices in Input-Output System . . . . . Some Problems of Input-Output Analysis . . . Theoretical Problems . . . . . . . . . . Operational Problems . . . . . . . . . . RESEARCH METHODS . . . . . . . . . . . . . . Review of Some Previous Reducing Techniques Which Are Not Appraised in this Study . . Modification of National Technical Coefficients . . . . . . . . . . . . . The Supply-Demand Pool Technique . . . . McMenamin—Haring Technique . . . . . . . Test of Accuracy of Regional Nonsurvey Techniques . . . . . . . . . . . . . . Reducing Techniques Appraised in this Study I O O O O O O O O O I I O C O O O O The Simple Location Quotient Technique . iv Page vii \l\lUlD-‘ |-‘ \O 10 10 ll 13 15 15 19 l9 19 21 24 24 26 IV. The Purchase Only Location Quotient Technique . . . . . . . . . . . . . The Cross Industry Quotient Technique The Modified Supply-Demand Pool Technique . . . . . . . . . . . . . The RAS Method . . . . . . . . . . . The Schaffer-Chu Iterative Procedure Income Multiplier Analysis . . . . . . . The Aggregate Income Multiplier Sectoral Income Multiplier . . The Type I Income Multiplier . The Type II Income Multiplier . Comparative Techniques . . . . . . . . . The Mean Absolute Percentage Deviation . . . . . . . . . . . . . The Mean of Relative Change . . . . . The Mean of Similarity Index . . . . The Structure of the Lansing Tri—County Region Input-Output Tables . . . . . . Sector Aggregations . . . . . . . . . . . RESULTS AND DISCUSSION . . . . . . . . . The Assumptions of the Six Reducing Techniques Compared . . . . . . . . . The Simple Location Quotient' . . . . The Purchase Only Location Quotient . The Cross Industry Quotient . . . . . The Modified Supply-Demand Pool . . . The RAS Method . . . . . . . . . . . The Schaffer-Chu Iterative Procedure Comparisons of Relative Efficiency of the Six Reducing Techniques in Estimating Direct Coefficients . . . . . . . . . . Income Multiplier Analysis . . . . . . . The Type I Income Multiplier . . . . The Type II Income Multiplier . . . . SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS Introduction . . . . . . . . . . . . . . Summary of this Study . . . . . . . . . . Relative Efficiency of the Six Reducing Techniques in Estimating the Direct Coefficients . . . . . . . . . . . Page 28 28 29 31 35 38 39 40 42 43 43 44 46 46 50 55 110 Page Relative Efficiency of the Six Reduc- ing Techniques in Estimating the Type I Income Multiplier . . . . . . lll Relative Efficiency of the Six Reduc- ing Techniques in Estimating the Type II Income Multiplier . . . . . . 112 Possible Sources of Errors . . . . . . 113 Conclusions and Recommendations . . . . . . 114 APPENDIX 0 O O O O O O O I O I O O O O O O 0 O O O 118 A, ESTIMATED DIRECT COEFFICIENTS GENERATED BY THE SIX REDUCING TECHNIQUES COMPARED TO THE TRUE DIRECT COEFFICIENTS COMPUTED FROM SURVEY DATA . . . . . . . . 118 B. THE RESULTS OF SECTORAL INCOME MULTIPLIER ANALYSES o o o o o o o o o o o o o o o o 144 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . 153 vi Table 1. 10. LIST OF TABLES Page Evaluation of Simulated Tables: Methods Ranked for Each Test . . . . . . . . . . . . . 25 The Study Area Aggregation Scheme . . . . . . 48 The Estimated Location Quotient (LQi) and Purchase Only Location Quotient (POLQi) . . . 60 The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (SLQ) . . . . . 61 The Statistical Characteristics of the ”True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (POLQ) . . . . . 65 The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (CIQ) . . . . . 67 The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (MSDP) . . . . . 71 The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (RAS) . . . . . 74 The Statistical Characteristics of the ”True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (SCM) . . . . . 76 Comparison of the Relative Efficiency of the Six Reducing Techniques . . . . . . . . . . . 81 vii Table 11. 12. 13. 14. 15. 16. 17. Evaluation of Estimated Tables: Reducing Techniques Ranked for Each Comparison Test . The Type I Income Multiplier of the Tri- County Region (Survey Method) Compared to the Type II Income Multiplier Produced by the Six Reducing Techniques, Appraised in this Study . . . . Comparison of the Relative Efficiency of the Six Reducing Techniques in Estimating the Type I Income Multiplier . Evaluation of Estimated Type I Income Mul- Reducing Techniques Ranked for tiplier: Each Test . . Type II Income Multipliers of the Tri-County Region (Survey Method) Compared to the Type II Income Multipliers Produced by Bach Reduc- 0 ing Technique Tested in this Study . Comparison of the Relative Efficiency of the Six Reducing Techniques in Estimating the Type II Income Multipliers . Evaluation of Estimated Type II Income Mul- Reducing Techniques Ranked for Each tiplier: Comparison Test viii Page 82 88 93 94 97 101 102 CHAPTER I INTRODUCTION Problem Setting In general, most state, regional and county planners are interested in local socio-economic evaluation and po- tential impacts of resource programs and plans. Informa- tion on costs, benefits, and other regional and environ- mental trade-offs are needed by decision makers, planners, and local participants, before recommending and implement- ing a plan. Indeed, this information cannot be just a localized national economic development display, but has to reflect the regional economy, including the interindus- try and intraindustry flow of goods and services. Reli-\_ able information has to be developed so that alternative _ regional develOpment plans and strategies can be prepared. and analyzed. In general, two major kinds of information/ are necessary: reliable estimates of structural relation; ships in a regional economy and direct and indirect impacts of planned economic activities. New economic activities must be translated into shifts in final demand on an indus- try by industry basis. For this purpose, input—output models are very useful, because they can be used to estimate structural relationships in a regional economy and impacts of planned new economic activities. Wassily Leontief spent a number of years in construct- ing input-output tables for the U. 8. economy for 1919 and 1929. His results, which he used to analyze the working of American economy were published in 1941 and 1951.1 An 87 sector input-output model of the U.IS. economy for the year 1958, was published by the Office of Business Econom- ics (now Bureau of Economic Analysis) in 1964. The larg- est U. S. economy input-output model is the 370 sector table published by Bureau of Economic Analysis in 1969 for the year 1963. The newest national model was published in 1974 for 1967. "With the publication of these results, the number of comparable benchmark input-output tables is in- creased to four and the time span covered by these studies is extended to 20 years covering the period of 1947 to 1967."2 Leontief in his analyses focused his attention on intersectoral relations, and based his analysis upon gen- eral equilibrium theory. While the general equilibrium theory of Walras did not go beyond writing down a set of equations with symbolic coefficients, Leontief estimated lLeontief, Wassily. 1941, 1951. The Structure of American Economy. New York: Oxford University Press. 2U. S. Department of Commerce, Bureau of Economic Analysis, "The Input-Output Structure of the U. S. Economy: 1967," Survey of Current Business, LIV (February, 1974), 24-560 such a system with empirically obtained numerical coef- ficients. In his conception, the statistical data collec- ted fill in the "empty box" of the theory. Hypothetical production and consumption equations gain explicit meaning as soon as the algebraic symbols are replaced by observed numerical values. Once empirical foundations are thus es- tablished, the vague generalities of abstract theoretical statements acquire empirical significance. The endeavors to make general equilibrium theory correspond to economic reality led to linearization of economics. This, together with the development of aggrega- tion schemes wherein millions of households could be repre- sented by a single sector and hundreds of firms of an indus- try by another, and so on, made the system amenable to an- alysis by use of matrix algebra. And Leontief used it with a telling effect. Essentially, input-output is a method of analysis that takes advantage of the relatively stable pattern of the goods and services in the short run, among elements of the economy to bring a more detailed statistical picture of the system into the range of manipulation by economic theory. Input-output analysis is applied intensively in the study of regional economy. An important reason for this is that input-output techniques can be implemented empir- ically in a field where data shortages and underdeveloped theoretical constructs restrict the scope for hard empirical research. Moreover, it provides virtually the sole avenue of escape from partial equilibrium analysis.1 The other general equilibrium theories available, Walrasian general equilibrium, Neo-Keynesian interregional macroeconomics, may be a little more satisfying theoretically but they are much more difficult to apply empirically. In particular, interregional trade flows are much easier to measure and make consistent with theory within an input-output frame- work. Moreover, some versions of interregional input- output models of which the gravity technique is typical, have the added advantage for analysis of the space economy that they take distance in the form of transportation cost explicitly into account as a relevant variable. There are at least four basic purposes of an input- output model: 1) It can provide a detailed description of a na- tional or regional economy by quantifying link- ages among sectors of the economy and the sources and origin of exports and imports, 2) Given a set of final demands (exogenous), total output in each industry and requirements for primary factors and resources can be determined, 3) The effects of change in final demands, arising in either the private or public sector, can be traced and predicted in detail, 4) Changes in production technology or relative prices can be incorporated by changing the technical coefficients of production. 1Richardson, Harry W. 1972. Input-Output and Re- gional Economics. Trowbridge, Wiltshire, England: Redwood Press Limited, 294 p. 2Lewis, W., Cris. 1971. Regional Economic Develop- ment: The Role of Water (Logan: Utah State University Foundation) p. 5). These advantages go a long way to offset the well known drawbacks--the restrictive assumptions, the high cost involved in obtaining data required, and practical obstacles in the way of operationalizing the dynamic models needed for long run regional planning. However, because of its useful- ness, regional economists in most cases must depend heavily upon input-output analysis. Their involvement can usually be traced to pragmatic grounds--that given the objective of their research, available data and research resources, in- put-output modeling is the most useful technique currently available. Objective of the Study The implementation of regional input-output models is invariably a compromise between the desire for theor- etical satisfaction and comprehensive data coverage, on the one hand, and the need to economise on time and resources on the other. Nonsurvey and minimum survey methods for constructing regional input-output tables are attractive to model builders because of the relatively smaller cost involved as compared with full survey methods. Yet, at- tempts at such nonsurvey and minimum survey techniques have not been successful. The distance between the survey and the nonsurvey direct coefficients matrices is quite large in absolute terms. In fact, until recently there were no accurate comparisons between tables constructed by survey and nonsurvey methods. Thus, a study of nonsurvey and minimum survey methods of constructing regional input-out- put tables would likely to contribute greatly to progress in regional economics. Input-output tables of the Lansing Tri-County Region (Eaton, Ingham and Clinton Counties) have been constructed by O'Donnell, 32 al., of Michigan State University in 1960,1 for the year of 1958 based upon primary data collected by a survey method. The purpose of this study is to construct nonsurvey and minimum survey input-output tables for the same region, for the year 1958, based upon the United States national input-output tables for 1958, by using six reducing techniques, i.e.: (1) The Simple Location Quotient, (2) The Purchase Only Location Quotient, (3) The Cross In- dustry Quotient, (4) The Modified Supply-Demand Pool, (5) The RAS Method, and (6) The Schaffer-Chu Iterative Procedure. The Type I and Type II sectoral income multipliers will be computed based upon the direct coefficients matrices produced by each reducing technique. The direct coefficients estimated by the nonsurvey and minimum survey techniques, as well as the Type I and Type II income multipliers generated by each technique will be compared to the "True" direct coefficients, the Type I and Type II multipliers computed by O'Donnell, 33 El: 1O'Donnell, John L., et al., 1960. Economic and Pop- ulagion Base Study of the Lansifig Tri-County Area: An In- ter-Industry Rélatién Analysis. Michigan State University, 31§"p. ‘ Comparisons will be performed by using three tests: (1) The Mean Absolute Percentage Deviation, (2) The Mean of Relative Change, and (3) The Mean of Similarity Index. Limitations of the Study The results of this study are applicable only to the Lansing Tri-County Region. Although the techniques ap- praised here may be applied elsewhere, extension of speci— fic quantitative results is not warranted. The empirically derived input-output model of the Tri- County Region is a regional version of the traditional Leontief open, static input-output model, and thus the non- survey and minimum survey techniques tested in this study produced input-output model which suffers from all problems and limitations of the Leontief open, static input-output model. The basic assumption of all techniques is that the national direct coefficients assumed appropriate at the regional level and that the regional direct coefficients differ from the national direct coefficients to the extent that goods and services are imported from, or exported to other regions. This assumption clearly implies the con- straints that the regional direct coefficients must always. be less than or equal to the national coefficients. 9112.12.22 Background information on previous related studies is presented in Chapter II. Research methodology is described in Chapter III. Results of this study are dis- cussed in Chapter IV. Finally, Chapter V provides a sum- mary, conclusions, and recommendations for further re- search on the nonsurvey and minimum survey input-output techniques. CHAPTER II LITERATURE REVIEW The Theoretical Foundations of Input-Output Analysis_ The theoretical foundations and problems of input- output analysis are discussed briefly in this chapter, to facilitate presentation and discussion of nonsurvey and minimum survey techniques in Chapter IV. The more elabor- ate discussions on this topic can be found in Richardson (1972),1 and Miernyk (1965)2. The input-output method is an adaptation of general equilibrium theory to the empirical study of the quantita- tive interdependence between interrelatedieconomic activi- ties. Leontief’s idea was as follows: An economy consists of a large number of consumers and producers who conduct among themselves transactions--sales and purchases of goods. The interdependence between the individual industries of the given system may be described by a set of linear equa- tions. Coefficients of these equations must be determined lRichardson, Harry W. 1972. Inputsgutput and Re- gional Economics. Trowbridge, Wiltshire, England: Redwood Press Limited[—294 p. 2Miernyk, W. H. 1965. The Elements of Input-Out— put Analysis. New York: Random House. 10 empirically. In the analysis of structural characteris- tics of an economy they usually are derived from a so- called statistical input-output table. Transactions Matrix Assume that one industry produces one commodity, there is no joint production. Hence, the number of indus- tries is the same as the number of commodities. Assume further that no changes occur in inventories of the commod- ities. By definition: jglxij + xj = x1 , i = 1,2,...n... (1) where: xij = intermediate demand for the ith industry's output by the jth industry's output; i = 1,2,...n. xj = final demand by consumers for the ith indus- try's output; j = 1,2,...n. xi = Gross output of industry i; i = 1,2,...n. The [xij] form a transaction matrix of processing sectors. Input Coefficients The static input output model assumes constant pro- portions among inputs and outputs for each industry. This production relationship rules out factor substitutability. Ratios between input from each industry and the total outlay, aij = xij/Xj ; 1,3 = 1,2,...n. (2) 11 are called input or direct coefficients and can be arranged as follows: l 2 ............ ...... n 1 all a12 . ......... ........ a1n 2 a21 a22 ............... ... a2n n an1 an2 ann a01 a02 aOn It should be noted that all of these coefficients are non- negative. Gross Output Needed to Produce Given Final Demands_ A model to compute gross output needed to produce given final demands is needed in constructing an input- output table as well as in forecasting. ‘The model can be formulated as follows: From (2), xij = aij - Xj; i,j = 1,2,...n Substituting (3) into (1), Z a..°X. + x. = X.; i,j = 1,2, n =1 13 3 J 1 j (3) ...n (4) 12 Let: a = input coefficient matrix = [aij] E = column vector of xi; i = 1,2,...n R = column vector of x.; i = 1,2,...n 1 Then in matrix notation, (4) can be written as follows: AX + i = Y; (5) Therefore: TIT—37;? = i If TI_:_AT is nonsinglular, then, if = Wk; (6) In formula (6), the (i,j) element of the Leontief inverse (I - A5 1 indicates by how much the gross output of the ith industry has to be changed corresponding to a_unit increase th or decrease in the j industry‘s final demand. In this sense, each element of the Leontief inverse can be called multisector multiplier. In input-output analysis the mul- tiplier represents the propagation effects through inter- mediate demand. For Example, _ 1 x1 ’ v (311x1 + lexz where v is the determinant of (I - A) and Bij are cofactors of the (i,l) elements of (I - A). The rate of change of th + ... + Bnlxn) (7) the first industry's gross output with respect to the i industry's final demand is: 3X1 Bil ___ = ___ (8) 13 Feasible Outputs In input-output analysis, output of the economy is limited by available primary factors. For an output vector X, the total labor needs are: n n ' _1 x0 = jilxoj = ji1a0jxj = a0 (I - A) x (9) where: aoj = the employment coefficients that is the ratio between the labor input and output a0 = column vector of aoj; j = 1,2,...n Xj = total outlay Prime indicates transpose of the matrix or vector. The Prices in Input-Output Systems Let the prices of n commodities be pi; the price of labor be w, and the total labor needs for sector j (j = 1,2,...n) are xO then for each commodity, jr + wx . = p.X. (10) 13' 03 J 3 substituting (3) and (4) in (10), n 2 .a..X. + wa .X. = .X. (11 i=1p1 1] 3 OJ 3 p3 3 ) Therefore: n iilpiaij + waoj = pj (l2) 14 Let: p = column vector of pi; i = 1,2,...n. Then in matrix notation (12) is equivalent to p'A + waO = E (13) That is, p' (I-A)=vT6 (14) 5' = wag, (I - A) l (15) Thus, ceteris paribus, given the wage rate, the prices of the commodities are determined by the input coefficients. Post multiply (14) through by X, obtaining, 5' ___—(I - 33.)")? = w—aoi (16) and premultiply (7) through p', obtaining 5' 73—7753? = 5'? (17) Expressions (l6) and (17) imply that, (7:363? = 5'; (18) Expression (18) states that income of the factor, labor, equals expenditure on final goods. Assuming constant returns to scale, the relative prices of the commodities are determined by the tech- nological condition represented by A, independently of output levels of these commodities. 15 Some Problems of Input-Output Analysis Theoretical Problems Input-output models incorporate the basic assumptions of the general Leontief model. Theoretically, each commod- ity is supplied by a single industry or sector of produc- tion, only one process is used for producing each commodity and each sector has only one primary output. There are, therefore, no joint products. Inputs purchased by each sector are a function only of the output of that sector. However, the translation into empirical tool involved lump- ing together only those plants with similar output and in- put structures. The practical solution is to group pro- cesses and products which differ in some respect but which behave sufficiently uniformly to be used as a basis for aggregation. The emphasis on linear production relationships also creates some problems. The essence of the Leontief model is the technological relationship that purchases of any sector (except final demand) from any other sector depend, via a linear production function, upon the level of output of the purchasing sector. The constant and linear produc— tion function assumption solves some kinds of difficulties, for instance it eliminates factor substitution and econom- ies of scale, but creates others. Time is missing, yet the purchase of inputs by one industry to make goods to sell to other industries implies a period analysis. Moreover, the notion of a linear production function is not very 16 meaningful in many nonindustrial sectors such as agricul- ture, services sectors and the government sectors, where production functions are usually not linear. In 1951, Samuelsonl showed how the assumption of linear relation- ship and no factor substitution are not as rigid as first appears. The Samuelson theorem as it is called, assumes that each industry produces only one commodity and that each industry uses only one scarce primary factor which is homogeneous for all firms in the industries. If it is further assumed that there are constant returns to scale, Samuelson found that even if each firm had a wide choice of alternative production processes, it is compatible with overall efficiency for each firm to use only one of the processes available. Extending input-output analysis to an interregional setting creates more problems, because it is necessary to make additional assumptions. A regional interindustry analysis must not only assume stable direct coefficients but also stable trading coefficients, so that repercussions of autonomous changes are not distorted by geographical differences in the location of final demand or in sources of supply. This means that all consuming industries ab- sorb the domestic output of any given type of good and service combined in a fixed pr0portion with a certain 1Samuelson, P. A. 1951. Abstract of a Theorem Con- cerning Substitutability in Open Leontief Model. In: T. C. Koopmans (ed.) Activity Analysis in Production and Allo- cation. John Wiley, pp. BIZ—146T 17 type of its competitive imports, the proportion itself being determined by the ratio of the total imports of the particular type of goods to their total supply. This as- sumption freezes the spatial structure of each region, and implies fixed regional supply areas for each and every consuming industry using a particular input. This assump- tion is restrictive since in reality the supply patterns of individual sectors for a given input may vary widely because of vertical integration, competition, different locations within a given region, entry of new firms, exit of firms or institutional ties. The interregional input-output analysis also implies comparative stability in relative prices between regions and no changes in interregional competitiveness in the supply of particular commodities. This implication is restrictive since in reality, there is probably even less justification for expecting stability in imports coeffic- ients than in direct coefficients. Finally, interregional input-output analysis as- sumes that imports enter a region only through the matrix of interdependence. Imports of a commodity are therefore considered as inputs and are fed directly into producing sectors, they enter final demand only indirectly as part of a regional industry domestic output. In some cases, it may be unrealistic to assume that imports enter only the producing sectors as inputs and never go directly to final consumption; for example, if a large consuming center is 18 located near the supplying region's boundary while the production plants of a sector are located at the maximum distance from that boundary, we would expect to find sup- plies of finished goods being sent directly to the point of consumption rather than the commodity being exported as an input to the distant plants, converted into final output and returned to the consumption center. Operational Problems A major drawback to the widespread application of input-output models at the urban and regional levels is a shortage of the requisite data. The successful implementa- tion of an input—output model demands an extensive data set which few other models need. Data are rarely published in a form directly useful for regional input-output stud- .ies, so the local analyst must collect all or some of the data through empirical survey, or can attempt to produce an inputdoutput table from the available published statis- tics. The former option will clearly be-the most expensive, the latter the least accurate, but often the most attrac- tive from the cost point of view. It should be noted that although survey based input-output tables are probably more accurate than the nonsurvey or minimum survey table, the former are frequently out of date when published, since research workers rarely pay for the cost of continu- ous updating. Thus, in most cases, nonsurvey tables may have to be based upon out-of—date input-output tables. CHAPTER III RESEARCH METHODS This chapter consists of six parts. The first is a brief review of some previous reducing techniques which are not appraised in this study. In the second part, the pro- cedure and assumptions of the six reducing techniques appraised in this study will be discussed. In the third part, the procedure and assumptions of income multiplier analysis are outlined. In the fourth part comparative techniques employed in this study are presented. Part five contains a description of the structure of the Lansing Tri-County Region input-output tables. Finally, in the sixth part, the procedure and assumptions of sector aggre- gations adepted in this study are discussed. Review of Some Previous Reducing Techniques Which Are Not AppréiSed in this Study Modification of National Technical Coefficients These techniques involve the modification of national coefficients as a result of detailed knowledge of certain local characteristics obtained either from an empirical survey or from published data. Into this cateogry can be placed the work of Isard and Kuenne on Greater New York- 19 20 Philadelphia,1 and the work of Moore and Petersen on Utah.2 They obtained a crude transaction table by using national direct coefficients to obtain sector columns from control totals. They then adjusted the row and column distribution for each industry in the light of regional production pro- cesses, marketing practices and product mix. Although in some respect this method could result in a more realistic regional input—output matrix, the subjec- tivity involved when changes are made in individual cells makes it difficult to test the reliability of the table in the absence of a survey based input-output table. The only other possible evaluative test for this technique is to assess the forecasting power of the table. The Supply-Demand Pool Technique Actually, this approach involves extending balance of trade computations to construct a regional input-output table. Commodity balances for each industry i is the dif- ference between input requirements and locally produced supply. If the commodity balance is positive, the national production coefficient can be substituted for the regional coefficient, sets imports to zero, and the surplus is assumed equal to export. If the commodity balance is 1Isard, W., and R. E. Kuenne. 1953. The Impact of Steel upon the Greater New York-Philadelphia Urban Indus- trial Region. Review of Economics and Statistics 35:289—301. 2Moore, F. T., and J. W., Petersen. 1955. Regional Analysis: An Interindustry Model of Utah. Review of Econ- omics and Statistics 37:363-383. 21 negative, exports are set at zero, imports are calculated as the difference between regional input requirements and locally available requirement plus imports for final de- mand and regional coefficients are estimated as: a.. = A..(x./r.) 13 13 1 1 where aij is regional direct coefficient, A.. is the na- 1] tional direct coefficient, xi is the regional output of in- dustry i, and ri represents total regional requirements (for inputs and final demand) of product 1. Thus, the re- gional direct coefficient is obtained by multiplying the national direct coefficient by the ratio of regional sup- ply to regional demand of producing sector i.- This pro- cedure allocates local production, where adequate, to meet local needs. Where the local output is inadequate, to meet local needs, however, the procedure allocates to each pur- chasing industry j its share of regional output 1, based on the needs of the purchasing industry itself to total needs for output 1. McMeanamin-Haring»Technique In 1974, McMenamin and Haring proposed a new method for estimating regional input-output tables.1 This method is simple in application and cost effective in that it re- quires little input data. It is only applicable to the problems of estimating a regional table at one date (year 1) lMcMenamin, D. G., and J. E. Haring. 1974. An Appraisal of Nonsurvey Techniques for Estimating Regional Input-output Models. Journal of Regional Science 14(2): 191-206. 22 based on similar table from an earlier point in time (year 0). The technique adjusts the regional table at year 0 for changes in price, the effect of substitution and the effect of fabrication which have taken place be- tween year 0 and year 1. First of all it is appropriate to introduce some notations which are used in the discus- sion of this technique, X = xij = regional input-output table for year 0 Y = yi = vector of total gross output for year 1 Z = zj = vector of total gross outlay for year 1 X3 = xt, ij = the regional input-output table after t/2 iterations i = 1,2,...m = the number of rows, including the payments sectors, and j = 1,2,...n = the number of columns, including the final demand sector. First of all this technique adjusts the regional input- output (year 0) for changes in relative prices, giving an estimated table X*. The iterative procedure begins by constraining the row sums of the gross flows (sales) in X* to the associated total gross output: The column sums of these row constrained gross flows (purchases) are further constrained to the total gross outlay: 23 n These column-constrained gross flows can be constrained to the total gross output. The technique alternates in this fashion until in iteration number t/2, n xt-1,ij - (Yi/ kilxt'z’ik) (xt-Z'ij) m xt' ij = (zj/ kilxt~l'kj) (xt-l'ij) The process is continued until the vectors of row and col- umn totals for the estimated matrix have converged to within e of the total gross output and total gross outlay, The matrix thus obtained, X is the estimated regional t' input-output table for year 1. The assumption behind the method are exactly par- allel to those RAS processes, and thus it is subject to the aggregation problem and uniform substitution effect. How- ever, it avoids the problem of having to estimate the total intermediate output and total intermediate input vectors. The only data needed for the procedure are an inputéout- put table for year 0 and total gross outlay vectors for year 1. 24 Test of Accurapy of Regional Nonsurvey Techniques Morrison and Smith1 compare the results of the appli- cation of several regional nonsurvey input-output tables in a consistent way with a survey based input-output model for the city of Peterborough, England. Their results are tabulated in Table 1. It is apparent from the table that the RAS method produces a superior simulation when judged by some of the measure distances. Of course, this is not entirely sur- prising given that the technique employs certain amounts of survey material. Some interesting features do emerge, for example, the Simple Location Quotient technique emerges as being superior to the Purchase Only Location Quotient on all five tests. The Cross Industry Quotient approach produced the poorest simulations on three most reliable tests and ranked seventh out of eight on the other two. It is perhaps a little surprising that the simplest of the tested methods (SLQ) emerges, on the whole, as the best of the purely nonsurvey approaches. Reducinngechniques Appraised in this Study In this part the procedure and assumptions of the six reducing techniques tested in this study are discussed. lMorrison, W. I., and P. Smith. 1974. Nonsurvey Input-Output Techniques at the Small Area Level: An Eval- uation. Journal of Regional Science l4(l):l-l4). 25 Table 1. Evaluation of Simulated Tables: Methods Ranked for Each Test* Test Mean . Mean Infor-- . Rank Absolute nggiiizion Similarity mation chire Difference . Index Content q 1ent 1. RAS RAS RAS RAS RAS 2. SLQ SDP SLQ SLQ CMOD 3. POLQ SLQ POLQ POLQ SLQ 4. CMOD POLQ SDP CMOD POLQ 5. SDP CMOD CMOD CILQ CILQ 6. CILQ CILQ CILQ SDP SDP *Source: Morrison, W. I., and P. Smith. 1974. Nonsurvey Input-Output Techniques at Small Area Level: An Evaluation. Journal of Regional Science.l4(l):1-14) Key initials of nonsurvey methods are as follow: SLQ: POLQ: CILQ: CMOD: SDP: RAS: Simple Location Quotient Purchase Only Location Quotient Cross Industry Location Quotient Modified Cross Industry Quotient Supply—Demand Pool RAS 26 The Simple Location Quotient Technique The Location Quotient is a measure comparing rela- tive importance of an industry in a region and its rela- tive importance in the nation. Thus, for industry i, where xi represents the regional output of industry 1, x is the total regional output, X. 1 is the national output of industry i and x is the total national output. If LQi :11, we assume that local production is ade- quate to supply local needs, then we may set aij Aij' where aij is regional direct coefficient defined as Xij/Xj' and Aij is the national direct coefficient defined as xij/Xj' Knowing regional industry outputs xj and hav- ing established aij’ we may compute regional interindustry flow as: Since, Aij = aij' d A.. = X.. X. an 13 13/ 3 then, xij = xij(xj/Xj) If the remainder of local final demand is known, as the region's share of national consumption, investment and gov- ernment expenditure vector (yif = Yif - x/X), where f in- dicates one of t final demand columns. Then, the exports 27 of industry 1 may be computed as a residual: t n I x.. - Z yif 3 e. = x. - 13 f l 1 If LQi < l, we assume that regional production is inadequate to satisfy local needs, then we may set a.. = 1] LQi-Aij, and regional gross flow as: xi: = 313 *3 Since aij = LQi Aij’ and Aij = Xij/xj' then x.. = x.. L . x. X. 13 13 Ql( 3/ 3) Imports of product i (mij) are computed as the amount necessary to satisfy production requirements, mij : Aij xj - xij The procedure just outlined empirically is grossly deficient. There is no guarantee, for example, that local production is adequate to satisfy local needs when LQi Z 1, nor that local production is inadequate to supply local needs when LQi < 1. Thus, because of the problems of these assumptions, to insure success in using the Simple Loca- tion Quotient technique, the local industry structure must resemble the national structure, this requirement is seldom met. 28 The Purchase Onlprocation gpotient Technique Due to problems encountered in using the Simple Loca- tion Quotient technique, Charles Tiebout suggested a modif- ication of that technique for the CONSAD Corporation to use 1 in constructing state input-output models. The Purchase Only Location Quotient may be defined as: ! xi/x LQi = x.7x' 1 where the prime indicates that summation includes only the outputs of those industries which purchase from industry i. Formulations for computing regional direct coefficients are not affected by the modification. The difference in these approaches lies in the values of the location quo- tients and thus in the determination of which industries are export producing andwhich are not. We cannot predict whether the Simple Location Quotient approach or the Pur- chase Only Location Quotient approach will yield the larger regional coefficients. The Cross Industpy Quotient Technique This quotient compares the proportion of national output of selling industry 1 in the region to that for purchasing industry j, 1CONSAD Corporation. 1967. Regional Federal Pro- curement Study. United States Department of Commerce Contract. 29 x./X. CIQ..= 1 1 ij xj7Xj If CIQ.. > 1, a.. = A.. for cell ij. Since the output of 13 —- ij 1] industry i is larger than that of industry j in the region relative to the nation, we assume that local industry 1 can provide all of the output required by local industry j. The reason is, if industry 1 in the region is producing a larger percentage of the national economy's product 1 than regional industry j is producing its product, it is likely that industry j can obtain all the product i it needs within the region. If CIQij < l, aij = CIQij’Aij. The regional direct coefficients becomes the national dis- tribution coefficient for industry i weighted by the ratio of the regional size of the selling industry to that of the purchasing industry. The Modified SupplyrDemand Pool Technique The pool technique has been modified in several ways. Kokat suggested an alternative which adjusts the Supply- Demand Pool technique for the case in which regional final demands are predetermined.l The Supply-Demand Pool is not affected if commodity balances are positive. The dif- ference arises when commodity balances are negative. Im- ports are assumed to enter the region as inputs but never for final demand, and exports are set at zero. lKokat, R. G. 1966. The Ecopomic Compgpent of a Regional Socioeconomic Model. IBM Technical Report 17-210, IBM, Inc. Advanced System Development Division. 30 The procedure in this approach is as follows: Com- pute input requirements based on national technology and estimates of local output, Given the final demand matrix for the region (Yif) compute total regional demand for goods excluding exports (ei), s t r. = X r.. + 2 y- i j 13 f if Compute commodity balances (bi)' where xi is total gross output of industry 1. Where bi is positive, substitute the national direct coef- ficient for regional coefficients, set imports at zero and compute exports, aij Aij x.. r.. 13 .1] mij = 0 ei = (xij/xi) bi Where bi is negative, compute imports (mij)’ set exports at zero and compute regional flows and-coefficients, 31 S mij = (rij/g rij) ’ b1 mij = (rij/(ri ’ Y1) (r1 ’ xi) where yi is total final demand sector which can be treated as a priori assumption. Regional flows are computed as residuals, .. .. m.. 13 13 13 or Xij rij (Xi - Yi)/(ri - Yi) This expression indicates that to each purchasing industry j we allocate its share of regional output i available to producers, based upon the needs of purchasing industry it- self relative to the needs of all purchasing industries. gag Stone and Brown,1 developed a method of estimating an input-output table for a certain date (year 1) , from a table constructed for an earlier date (year 0). This technique, called the RAS method, obtains the estimated table through adjustment of the direct coefficient, aij' of the base year table to account for changes which have been taken place between these two periods. These changes are of the following three types: (1) changes in the relative level lStone, R., and A. Brown. 1962. A computgble Model of Economic Growth (A Progpamme for Growth 1). London: Chapman and HdlIL 32 of prices, (2) changes in the degree to which commodity i has uniformly been substituted for or replaced by other intermediate inputs (called the substitution effect), and (3) changes in the degree to which intermediate inputs have uniformly increased or decreased in weight in the fab- rication of commodity j. In this study the RAS method has been adopted to the problem of estimating regional input-output tables from the national table. The adjustment process forms an iter- ative procedure which has been shown to converge under cer- tain conditions. The method requires the following data as input: (1) the national direct coefficient matrix, (2) the regional total gross output vector, (3) the regional total intermediate input and intermediate output vectors. Stone and Brown obtained the total intermediate input vec- tor by subtracting the value added vector and import vec- tor from the total gross output vector. They obtained the-total intermediate output vector as the difference be- tween the total gross output vector and the total final demand vector. Assumptions of the technique can be summarized as follows: (1) Price differences operate uniformly along rows, whenever there is difference in the average price of the products of a sector, it is charged in the same proportion to all users. (2) Whenever there is a substitution of one product 33 for another due to difference in demand or in- dustry mix, it affects all users to the same extent. (3) Wherever there is a change in the degree of fabrication, it unifromly affects all produc- tive processes. In the skeletal procedure below, the notational con- ventions are that the lower case letters refer to vectors, upper case letters refer to matrices, hatted letters de- note diagonalized vectors, subscripts refer to successive estimates, a prime indicates a transposed matrix, and i is the identity vector. A is the matrix of national input- output coefficients, and u, v, and x are vectors of re- gional intermediate output, intermediate input and gross output respectively. In the first stage of the procedure, the national input-output matrix is treated as a first estimate of the regional table and is combined with the vector of regional gross output to yield an estimated vector of intermediate outputs. Thus, u1 = Alx The matrix is then adjusted to conform with row constraint (u) . A2 = uul Al The matrix A2 is then used to estimate a vector of 34 intermediate inputs, = ' I v1 ixA2 The matrix is then adjusted to conform with the column constraint (v), A3 = szv The matrix A3 is then substituted into 111 = Alx, and this process is repeated until the matrix converges to a state in which both row and column constraints are fulfilled. The RAS method was tested in national level (Belgian input-output table) by Paelinck and Waelbroeck.1 It seems that the RAS method tends not to work very well, probably because the method spreads any changes evenly along rows and down columns and such spreading may be inappropriate. Czamanski and Malizia, adopted the RAS method to the problem of estimating regional input-output tables from national tables.2 This study also measured the accuracy of the estimated tables through the use of a summary of the "de- viations between estimated and true regional input-output coefficients.“ The authors concluded It appears that although national input-output tables cannot be used for the purposes of regional studies. without considerable adjustments, acceptable results can be achieved by the methods tried on the Washing- ton State table. lRichardson, Harry W. 1972. Op Cit., p. 175. 2Czamanski, S., and E. E. Malizia. 1969. Applicabil- ity and Limitations in the use of National Input-Output Tables for Regional Studies. Papers and Proceedings, Re- gional Science Association 23:65-77, p. 77. 3S Czamanski and Malizia are concentrating on obtaining the interindustry direct coefficient. However, this is only part of the bigger problem of estimating an entire regional input-output table, which includes estimating final demand by sector, value added and imports. Particularly trouble- some in regional tables is the generation of imports and exports by industry. Yet Czamanski and Malizia have ig- nored these problems and have in fact assumed these sector levels to be known. Thus, the accuracy tests performed by them give results under the best possible conditions and thus place a lower limit on the possible errors generated by the method. Schaffer-Chu Iterative Procedure Schaffer and Chul have constructed input-output tables from national tables using a technique that they called the iterative simulation technique. The Regional Input-Output Table (RIOT) simulator not only assumes that the national direct coefficient applies, but also attempts to distribute local production according to both the national sales pattern and local needs. Simply stated, they compute the required inputs (rij) for producing estimated regional output (xj) for each industry and estimate local final demand rij as a lSchaffer, W. A., and Kong Chu. 1969. Nonsurvey Techniques for Constructing Regional Interindustry Models. Papers and Proceedings, Regional Science Associations 23:83-101. 36 proportion of national final demand. They then allocate local sales for each industry, basing the initial step on the national sales distribution pattern. Next, they use an iterative process to reallocate local sales, row by row until local production and consumption needs are satisfied to the extent possible. If local output is in excess of locally required sales for an industry, the ex- ports are computed as a remainder. They record imports in a matrix as the positive difference, for each cell in the regional transactions matrix between local production re- quirements and the local output available to meet these requirements. This completes the computations for a re- gional gross flow table. In a more formal expression the procedure may be de- scribed as follows: (1) Compute purchases required to produce regional output (rij), based upon national direct coef- ficients matrix (Aij), r.. = x.-A.. 1] J 13 (2) Distribute regional sales (dij) initially by national distribution, dij = xi (Xij/xi) (3) Compare requirements and allocations to deter- mine surplus allocation to cells (zij) and con- struct for each row i a pool of surplus 37 available for allocation (POOLi) and pool of needed reallocations (NEEDSi). The final de- mand matrix (zyif) is treated the same way; therefore, POOLi = sum of all pOSitive zij and zyif NEEDSi = sum of all negative zij and zyif (4) Allocate sales for industries with exportable surpluses POOLi > —NEEDSi, x.. = r.. 13 1] yif Cir ei = POOLi + NEEDSi Each buying industry receives all its needs of product i and the remainder is exported. (5) Reallocate local sales of industries with out- puts insufficient to meet local needs (0 < POOLi < -NEEDS.) -' 1 zij is positive or zero, xij = rij When zij is smaller than zero, compute 38 xij = dij + POOLi (dij/Xi) This procedure is iterated until POOLi goes to zero. When POOLi is equal to zero there is nothing to add to dij; thus, the iterations cease. This spreads the surplus of local output among industries on the basis of relative need. In general, the Schaffer-Chu Iterative procedure differs from the Supply-Demand Pool technique in attempting to follow the national sales pattern to distribute local output and in reallocating local sales from one cell to another as necessary to best satisfy local needs. It thus appears to represent the feasible maximum of local trade. Income Multiplier Analysis The propagation effects of change in the level of expenditure upon income in a region can be estimated by a multiplier. The concept of the multiplier was first in- troduced into economic theory by R. Kahn in his article "The Relation of Home Investment to Unemployment" pub- lished in the Economic Journal in June 1931. The multi- plier concept was elaborated, adapted and made an import- ant part of his theory of income and employment by J. M. Keynes.l Both Keynes and Kahn dealt primarily with lKeynes, J. M. 1935. The General Theory of Emplpy- ment, Interest and Money. Harcourt, Brace ahd Co., New York, pp. 112-131. 39 aggregate multipliers. In general, they were interested in measuring the total income and employment changes in a national economy resulting from exogenous changes in in- vestment. Of course, such aggregate economic analyses have become important guides to national economic policy decisions. Aggregate economic analysis is sufficient for many purposes. But if we are interested in the impact of changes in a single sector upon all other sectors, it is necessary to go beyond the aggregative analysis. The advantage of the input-output model is that it permits the analyst to focus attention on individual sectors of the economy. Given the available data--in the form of production and consumption functions--it is possible to compute income multipliers for the sectors defined by an input-output table. The Aggregate Income Multiplier The general expression for the Keynesian income multiplier is: K=1-199- AY where: K = the multiplier AC = changes in consumption AY = changes in income The above equation indicates that aggregate income 4O multiplier is a function of the marginal prOpensity to consume. It is assumed that with each increment of income there would be an increase in consumer spending, but that some fraction of the increment to income would also be saved. Thus, the marginal propensity to consume would al- ways be less than one. Keynes made the further assumption that as income continued to rise the propensity to consume would become smaller. In this formulation of the multiplier the only leakage is that to savings. It also indicates that each injection of new income will produce successive rounds of consumer spending. Sectoral Income Multiplier The technique for estimating a sectoral income multi- plier is more complex than procedure described above. This is true at the level of national economy, moreover the difficulties are increased when sectoral income multipliers are estimated for region, as have been done in this study. The reasons are: (a) the requirement that a consumption function for each regional sector must be computed, (b) the problem of additional leakages which make estima- tion of sectoral consumption rather difficult. Sectoral income multipliers have been calculated in some earlier input-output studies. However, the Moore and Petersen,1 and Hirsch2 studies were pioneering studies lMoore, F. T., and J. W. Petersen. 1955. Regional Analysis: An Iterindustry Model of Utah. The Review of Economics and Statistics. Vol. XXXVII, No. 4, pp. 368-383. 41 in regional input-output analysis, and they represent an improvement over earlier input-output studies. In these studies, however, only a limited number of sectoral con— sumption functions were used, and these are based upon national data. Thus, in neither case were the writers able to show the leakages in consumer spending with given increases in income and this imparted an upward bias to their regional income multipliers. Although this is re-' cognized by the author of this study, it was not possible to make the necessary improvements, because of lack of data. Two types of sectoral income multipliers have been computed for this study: i.e., (1) Type I Multiplier, and (2) Type II Multiplier. The Type I income multiplier is the ratio of direct and indirect to direct income change resulting from the delivery of one dollar to final demand by a given sector. Type II multipliers are the ratio of direct, indirect and induced to the direct in- come change, resulting from the delivery of one dollar to final demand by a given sector. Thus the Type I multipliers should be considered as first approximation, particularly for an "open" economy. The Type II multipliers, which are larger in every sector are more accurate estimates of the income changes pro- duced in the region by changes in final demand. In other 2Hirsch, W. Z. 1959. Interindustry Relation of A Metropolitan Area. The Review of Economics and Statis- tics. Vol. XLI, No. 4, pp. 360-369. 42 words, the Type II multipliers show the effects of suc- cessive rounds of consumer spending (including the induced effects) in addition to the direct and indirect effects of increases in sales to final demand. The procedures for computing these multipliers can be described as follows: The Type I Income Multipliers (l) (2) (3) (4) (5) Define, A a 20 by 20 matrix of direct coefficients, aij’ where the household sector is not in- cluded in the processing sector. » ll h a 21 by 21 matrix of direct coefficients, h 13' in the processing sector. a , where household industry is included Compute the Leontief inverse, (I - A)’1 where I is an identity matrix. Define Hr as the vector of direct income change, where Hr is the household row of matrix Ah. Calculate the direct and indirect income change as: (I - A)-l . Hr Calculate indirect income change as: Direct and indirect income change - Direct income change (see Appendix B, Table 1-7, column 4). 43 (6) Compute the Type I income multiplier as: Direct and indirect income change Direct income change The Type II Income Multiplier (1) Define: Ah = a 21 by 21 matrix of direct coefficients, agj, where household industry is included in processing sectors. (2) Compute Leontief inverse using the augmented matrix Ah, -1 where I is an identity matrix. (3) Define the direct, indirect and induced income change as the household column of (I - Ah)-l. (4) Compute Indirect and induced income change as: Indirect income change + Induced income change (see Appendix 7, Tables 1-7, column 7. (5) Compute the Type II income multiplier as: Direct, indirect, and induced income change Direct income dhange Comparative Techniques An important part of this study is to establish and select some objective means of evaluating the distance-- or relative distance--between matrices. In other words, it is desirable to have some consistent test of judging 44 which of the estimated coefficients matrices is closest to the equivalent survey based table. Indeed, this implies that the coefficient of the survey based table are assumed to be "true" (assume that no error exiSts). In this study three comparative techniques are used in evaluating the efficiency of each reducing technique. Those are:’ (1) The mean absolute percentage deviation, (2) The mean of relative change, and (3) The mean of the similarity index. First of all it is appropriate to introduce some notations which will be used in the discussion of this part, aij = direct coefficient for the survey table bij = direct coefficient for the estimated table n = number of industries in producing sector m = total number of direct coefficients. The Mean Absolute Percentage Deviation Define the absolute percentage deviation as: Iaij ' bijl dij — aij , for aij # O The mean of these absolute is thus, ij m 45 The standard deviation of these absolute percentage is: m n 2 m n 2 s = Z 2 di' - ( Z 2 di') i=1 j=l 3 i=1 j=l 3 This technique is not ideal, the diSadvantage is that it cannot handle a situation where aij is zero and bij is not zero--a situation which is not uncommon in this type of study. In this case, such elements are omitted in the calculation, and no credence is given to actual estimated values. Instead, the technique is used only as a tentative measure. The Mean of Relative Change Define the index of relative change as: aij - bi.| Rcij = 172(aij+hij) The mean of these indexes of relative change is thus, P45 Mb Rij = j=l i=1 RC m A rather unconventional feature of this mean is that it ranges from zero to two. The closer the value to zero, the better the estimates. 46 Mean of the Similaritnyndex The similarity index is defined as: The value of this index ranges from zero to one. Of course, the closer the value is to one the better the es- timates. The Structure of the Lansinngri-County Region Input-Output Tables In this part the structure of the Lansing Tri-County Region input-output tables is discussed briefly. The more elaborate discussion of this topic can be found in O'Donnel, a 21.1 The Lansing Tri-County region input-output tables consist of 24 intermediate demand or processing sectors and six final demand or autonomous sectors. The process- ing sectors are: (1) Livestock, Dairy and Poultry, (2) Crops, Vegetables, Fruits, and Nuts, (3) Food and Kindred Products, (4) Lumber and Furniture, (5) Printing lO'Donnell, John L. 1960. Economic and Population Base Study of the Lansing Tri-County Area: An Inter- Industry Relation Analysis. Michigan State University, 319 p. 47 and Publishing, (6) Chemicals and Allied Products, (7) Miscellaneous Nondurable Products, (8) Primary Metal, (9) Fabricated Metal Products, (10) Machinery including Electrical, (11) Motor Vehicles, (12) Miscellaneous Dur- able Products, (13) Electric Power and Gas, (14) Transpor- tation and Communication, (15) Wholesale Trade, (16) Retail Trade, (17) Finance and Insurance, (18) Real Estate and Rental, (19) PerSonal Services and Amusements, (20) Bus- iness Services, (21) Repair Services, (22) Medical and Other Professional Services, (23) Education and Non-Profit Organizations, and (24) Mining and Others. The final demand sectors are: (1) New and Mainten- ance Construction, (2) Non-Competitive Imports, (3) Federal Government, (4) State and Local Governments, (5) Gross Private Capital Formation, and (6) Households. The aggregation scheme adopted in this study, as well as the definition for each category will be presented in Table 2. It is not possible to make definite appraisal in general about the accuracy of the entries in the table. On the average 60 percent of the inputs of the processing sectors were obtained by direct survey method. Wherever possible the data in the table cover the calendar year of 1958, but in some cases some small and medium sized firms reported fiscal year of 1958 data. The numbers in the transaction table are in thou— sand dollars, each set of numbers in particular column 48 .mo uumm "um “mmooo HmwuumsocH pudendum ”on “coflmmm monsoolfiue "Ole "mmuoz mo mm.uo .em.oo He.oo.em o moooooum Hope: oooooeoooe .m oomm .momm .momm.oo .Homm .Homm .ommm .Nmmm .mmmmiflmmm .Hmm.oo mm.em m mambo: somehoo .e o~.uo .om~.oo .o-.od Hm.mH .e-.oo .o-.oo .e-)H-.oo .mH.eH.oH e mooooooo ofloouooueoz moooomeeoomez .e mm meow .Hemm .meuaemm .mmmm o~.m~.em o muoooooo ooeaaa ooo Hooeeoao .m mmfluumsocH me.oo .o-ae~ om m ooeaee oeo .ooeamefiooo .ooeoouoo .e mm mm.ud .e~.uo .-.H~.o~ o muooooum .mooos .uoosoq .ousuecuom .m mo.oo .moo .mmo .omeo.oo .mmeo .Heo .Hoo .omo .emo .Neo .Hmo .eeo .mo.oo moohouom one ooao.~oao .eao.od .mao .HHo ea.e.m.m m.~ .mooooooo sooamem Houooaooeooa ooaoo .m «no.oo No.6o .maoo .eHo .oo .mao H H mooooooo xooomo>ea one xooooooeq .H muouomm muouowm wauwe muouuom . moooo uHm emoa .m.: mmoa ous mmma suumooeH oz ll Ilaii‘l OI. mewcom sewummmumm< mou< mosum one .m oases 49 .mo ammo ”.um “mmmoo ammuum5psH oucocmum "UHm usoflmwm >DGDOUIHHB "one "mmuoz mva .mva .mva .eeH .mea .HeH .HmmH .HHMH o.m em mooeoo one moose: om Hmmm .om meowumNflcmmHO Damoumcoz .vm .Nm .om .Hmmu .Nmno hm mm.~m cam .mmofl>uwm HosOMDMODom .HtOflooz ma me me am . mooe>uom shadow me Hm.um .m>.um mu om mmuw>umm mmocwmom NH mm .mh.um .wn.um .m>.um .on.um mn.mh ma ucwEmmsE¢ pom anaemumm ma oe.ud Hoooom one .Hm .ee.oo .me.uo .eeuoo anion mH.eH oooomm Home .ooeooomoH .moocooem me mm.mm.um .nmnmm .Hm.um .om mm oH.mH mpmue Hfimuom paw oammoaocz «H we.» . .me.oo .Ne.oo .He.oo .ee.oo oo.mo ea :oeoooeeoeeoo oee compouooomeooe me . mmuw>umm me we ma humuflcmm oco .mmo .HOBOm Ofluuowam NH mm.vw om .ommiamm .mm (mo .omsmm NH mooooooo manouoa mooocoafloomez .HH Hum mm Ha mwauw£o> Houoz .OH mom .nmm1Hom .mm mmlmv ca . HMOflHuomHm mswosausH Sumsflcumz .m muouumm muouuow wauwe muouomm .m.o mmoe one mmoa soomoooe oz wmpou UHm nmmH Iioll .1 DIIIIIII" .tl.1||.|lilitl.. Iliallliiicl‘." I'll. ill!" H l I i l I l t o i I Gilli. t ill]: .. .-lilllll ii- ‘ 1‘1'II. nomssflucoov .m manme 50 represents the purchases or receipts of the sector name in the caption (at the tOp) from the sectors listed at the left side (in the stubs). Conversely, the set of numbers in a particular row represents the sales of the sector named in the stub to the sectors listed in the column cap- tions. Sector Aggregations Most of the practitioners of input-output analysis are aware of the importance of the aggregation problem and the fact that results of the analysis depend upon the aggre- gation procedures employed. The purpose of introducing the aggregation process into input-output analysis is to reduce the number of equations and unknowns. A user of input- output analysis often needs to reduce a given table to smaller size. A large tabel--say 100 by 100--is cumber- some for many purposes. It should be recognized, however, that if the results are to be useful, the system must be reduced to manageable size and yet still must describe significant relationships between many components of the aggregates. It is even possible that the particular infor- mation lost may be of greater importance than the more gen- eral information available in the solution made possible by aggregation. Thus, the aggregation decision involves a trade off between information lost due to aggregation and cost of data analysis. The sector aggregation decisions in this study are 51 based upon several considerations, i.e.: (l) (2) (3) The purpose of the study. The purpose of this study is to evaluate six nonsurvey and minimum survey reducing techniques for estimating the Lansing Tri-county region input-output table. In general, a very detailed input-output table is not necessary for testing reducing techniques, because results of this study cannot be used directly to guide policy making. The nature or sector detail of the base model input-output analysis in this study is based on a 87 sector model of national input-output for 1958. Working with a 87 by 87 matrix is not only cumbersome but also expensive. Cost of data analysis. Budget available for this study is very limited, and the cost of inverting a matrix rises exponentially with the size of the matrix. In this study, two base model matrices were used, (1) (2) A 87 by 87 matrix of 1958 National input—output table. A 24 by 24 matrix of 1958 Tri-county Region input-output table. The purpose of the aggregation procedure in this study is to reduce the size of both matrices to a 20 by 20 standard matrix (that is the size of matrix we used in 52 this study). The reason for adopting this aggregation scheme is that even if in general the national table is less aggregated than the regional table, for some specific sectors the regional table is more detailed (e.g., sectors number 22 and 23 of 1958 Tri-county region matrix). Thus, these sectors must be aggregated to match the corresponding national sectors. The entire sector aggregation scheme is shown in Table 2. As illustrated in Table 2, four aggregations have been performed in reducing the 24 by 24, 1958 Tri-county region matrix to a 20 by 20 standard matrix. The four aggregations are: (1) Sector number 2 and number 3 of the Tri-county matrix (the Crops, Vegetables, Fruits, and Nuts; and Food and Kindred Products respectively) have been aggregated into one sector and termed Other Agricultural, Fishery Products and Services. (2) The Wholesale and Retail Trade sectors (sector number 15 and 16 respectively) of 1958 Tri- county matrix have been aggregated into one sector, that is the Wholesale and Retail Trade sector (sector number 14 in the standard matrix). (3) The Finance and Insurance; and Real Estate and Rental sectors (sector numbers 17 and 18 re- spectively) of 1958 Tri-county matrix have been aggregated into one sector, that is the Finance, Insurance, Real Estate and Rental sector 53 (sector number 15 of the standard matrix). (4) Sector numbers 22 and 23 (Medical and Other Pro- fessional Services; and Education and Nonprofit Organizations sectors respectively) of 1958 Tri-county matrix have been aggregated into one sector--Medical, Educational and Nonprofit Organizations--that is sector number 19 in the standard matrix. In order to be compatible with these aggregations, the 87 by 87 endogenous sectors of 1958 national matrix have been aggregated, following guide lines as suggested by Cen- sus of Employment data. Before discussing results of this study, it is appro- priate to point out some assumptions made in preparing the study area input-output model from the national model. Assumption 1. Sector composition of the study area is similar to the national sector composition. Sectors are composed as a rule, of firms having sim- ilar input requirements. To the extent that a sector at the regional level contains firms with different input re- quirements from those in the nation, some aggregation er- rors may be introduced to the results of the study. The greater the aggregation, the less troublesome will be a violation of this assumption; but also the less detailed will be the conclusions, which can be drawn about a spec- ific industry. Assumption 2. The production function of any sector 54 i is similar in the study area as for in the nation. This implies that all inputs produce with the same degree of efficiency in the region as in the nation. It leaves cpen the possibility that some industries will have to import from outside the region. The reducing techniques tested in this study contain mechanisms to adjust the co- efficients for imports (if the region industry 1 is able to supply only half of the total requirements of other en- dogenous industries for product i, the remainder will be imported). CHAPTER IV RESULTS AND DISCUSSION This chapter consists of three parts. In the first part the assumptions of the six data reducing techniques appraised in this study will be discussed and compared. The second part will concern the relative efficience of each reducing technique in estimating the regional direct coefficients derived from the national tables. Finally, in the third part, the relative efficiency of each reduc- ing technique in estimating the value of the Type I and the Type II multipliers is evaluated. The six reducing techniques tested in this study em- ploy mechanical routines to adjust the direct coefficients of the United States model to reflect the economic struc- ture of the Tri-county region. The six methods differ in their data requirements. The first three methods, i.e.: the Simple Location Quo- tient, the Purchase Only Location Quotient, and the Cross Industry Quotient, require only the total gross output fig- ures for each industry in the study area. The Supply- Demand Pool and the Schaffer-Chu Iterative techniques re- quire the total gross output and the final demand data of 55 56 the study area which are estimated from the final demands of the United States model. The last technique, the RAS, requires not only the total gross output, but also the total intermediate input and total intermediate output of every industry in the study area. Each technique involves different costs, however, the RAS technique is the most costly because it employs a certain amount of survey data. The Assumptions of the Six Reducing Techniques Compared As may be expected, the relative efficiencies of reducing techniques is adjusting the national input-output model to reflect the economic structure of the Tri-county region differ from one another. These differences are due to differences in assumptions. Thus, it is important to discuss and compare the assumptions of each technique. This may facilitate the interpretation of the results of this study. The Simple Location Quotient Technique As has been pointed out earlier, the Location Quo- tient is the ratio between the percentage of total output supplied by industry i in the region, and the percentage of total output supplied by industry i in the nation. Thus, a location quotient of unity for industry 1 means that the region has its proportionate share of the entire production of industry in the nation economy judged in terms of total 57 regional production, relative to total national production. Then, if LQi Z 1, it is assumed that the region is producing more than its local consumption and has product available for exports. If LQi i 1, regional production is assumed to be inadequate to supply local requirements; so that imports are necessary. The reader must be aware of the limitations of these assumptions. These assumptions must be seriously qualified. Indeed, there is no guarantee that a surplus for exports will exist when LQi i 1, nor that local production is inadequate to supply local needs, in case LQi < l. The reasons are: (1) tastes and propensity to consume among households of the same type and income differ between region. In Southern Florida, little fuel is needed by households; in Michigan greater amounts. This means that for the fuel manufacturing sector, a location quotient of one for Southern Florida may involve exports of fuel, and for Michigan it will most likely involve major import of fuel oil; (2) income levels of households differ among regions. Families in Michigan consumed, per house- hold, more shoes than is the case in Georgia. Given this, a location quotient more than one in Michign for the shoe manufacturing industry could be consistent with imports of shoes, meanwhile, a location quotient below unity in Georgia may result in net exports; (3) production efficiency dif- fer among regions, and (4) industrial mixes vary consider- ably among regions. The shortcomings of these assumptions have far 58 reaching implications for the direct coefficients genera- ted by the Simple Location Quotient technique. It has been described earlier that if LQi 1 1, that the regional dir- ect coefficient is identical to the national direct co- efficient; and if LQi < l, the regional direct coefficient is equal to LQi times the national direct coefficient. Thus, the national direct coefficient can only be adjusted downward but not upward. As a consequence, this method cannot handle adequately a situation where a "true" re- gional direct coefficient is far larger than the correspond- ing national direct coefficient. The downward adjustment process in this procedure also has some drawbacks. To i11- ustrate this, we may define aij as the regional direct coefficient, and Aij as the national direct coefficients of row i. Even if aij is larger than Aij' in a case of LQi < 1, it follows that Aij must be adjusted further down- ward. Thus, the adjustment process is based upon "average variation" among rows and cannot handle variations within a row. This discussion suggests that the Simple Location Quotient technique is likely to give satisfactory results, that is, it generates regional direct coefficients that adequately represent the region's economic structure or sectoral linkages when the local industry structure closely resembles the national model structure. The preceeding discussions regarding the problems of the Simple Location Quotient technique does not mean to suggest that the Simple Location Quotient technique is 59 useless, but rather to point out some of its limitations, so that the reader is aware of the problems of this tech- nique. The location quotient for each industry in the Tri- county region has been calculated in this study and the results are presented in Table 3. As can be observed from Table 3, on the surface it would appear that only two industries--(l) Motor Vehicles, and (2) Medical, Educational Services and Nonprofit Or- ganizations--whose location quotients are more than unity, are exportrindustries, and the rest (18 industries) whose location quotients are less than unity are import indus- tires. This statement must be partially rejected. Indeed, the "Motor Vehicles" and the "Medical, Educational Ser- vices, and Nonprofit Organizations" are export industries. However as the survey based table revealed,l there are three more export industries-~(l) Other Agricultural, Fish- ery Products, and Services, (2) Chemical and Allied Pro- ducts, (3) Finances, Insurance, Real Estate and Rental—-out of the 18 industries, whose location quotients is less than unity. Thus, it may be concluded that, even if an indus- try's location quotient is smaller than unity, it is not necessarily an import industry. Based upon the location quotient in Table 3, the es- timated direct requirement per dollar of output were 1O'Donnel, John L., pp al., 1960, gp_cit., p. 256. 6O ucoeuoso :owumooq waso mmmnousm n «040m m ucmwuoso :oflumooq u Hana mowed. mmhmo. muwnuo use mcwcez .om mamah.a ~mmmm.a uneducaflcmmuo aflmoumsoz one mm0fl>uom stofluoospm .Hmoeomz .aH demon. Hamem. mooe>uom needom .mH Hhvom. mvmom. moufi>umm mmocemom .SH Hmmvm. mmoov. mucoEmm584 use mooe>umm Modemumm .wa mmmmv. mmmmo. Houcmm poo ououmm doom .mucmusmsH .moocmsflm .ma Hoomv. Hovmv. momma aeoumm one mammwaonz .oa omovm. omovm. coauoOflcoEEou use :ofluouuommcmua .ma Hmmmv. Hmmmvu mw0fi>umm >umuflcmm poo mow .uo3om uauuooam .NH «meow. ooohm. muosooum manmuso msomcmaawomflz .HH mowmm.~a mmowm.- moaoonm> Mono: .oa hmwom. nmomm. Hmofluuowam oceosaocfl whosenomz .m mmnom. momom. muooooum Houoz woum0flunmm .m mmmho. vmvoo. mamuoz SHMEMHm .n memos. momoo. whosoouo maoouoocoz msomsmaamomflz .o vmmma. ooama. muosooum omfiaad use HMOflEwno .m Namem. Namem. eoeooeooee oedema oee oeeaeeaooo .oeeoeeoo .e owmom. momam. :ofluosooud moooz .Hmnesq .musuwcuom .m mommo. oamom. mw0fl>uwm one muuopoum muwnmwm .Housuasoflum< umnuo .N mmamo. mmmvm. mausooum xooumo>eq one xuoumw>flq .H meoqom Heoq mosqescome oeoz Hooomm .oz .c0emom >UCDOUIHHB on» new ucwfiuoso :oflumooq waco ommnouom owmefluwm on» use ucmwuooo sofluoooq omumEfiumm one .m canoe 61 developed for the Tri-county region and the results are presented in Appendix A-l. Estimated regional direct coefficients are compared with "true" regional direct coefficients later in this chapter. However, at this stage, it is appropriate to com-; pare statistical characteristics of the estimated coeffici- ents to the true direct coefficients. The comparison is presented in Table 4. Table 4. The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (SLQ) Techniques Characteristics 1 2 DSM SLQ Mean .019 .009 Variance .002 .001 Standard deviation .050 .024 Kurtosis 31.249 62.153 Skewness 5.191 6.830 lDSM = Direct Survey Method 2SLQ = Simple Location Quotient As shown in Table 4, the Simple Location Quotient technique on the average, underestimates the regional dir- ect coefficients by around 50 percent. And, it follows that both variances and the standard deviation are 62 underestimated by 50 percent also. Indeed, this is not really surprising, because only two sectors--(l) Motor Vehicles, and (2) Medical, Educational Services, and Non- profit Organizations--out of 20 sectors, have location quo- tients greater than unity. Thus, the estimated direct co- efficients of these two sectors remained identical to the United States coefficients, since in these two sectors ade- quate local production is available to meet both proces- sing and final demands requirements. The estimated direct coefficients of the other 18 sectors were computed by ad- justing downward the corresponding national direct coef- ficient by the average of their location quotients. Kurtosis is that property of a distribution which ex- presses its relative peakedness. The greater the value of kurtosis, the more peaked will be its distribution. A normal distribution has a kurtosis value of 3, and is called mesokurtic. If the kurtosis is greater than 3, as in this case, the distribution is called leptokurtic. As seen in Table 4, the kurtosis of the estimated direct coefficients produced by the Simple Location Quotient is almost two times as large as the kurtosis of the "true" regional dir- ect coefficients. Skewness means lack of symetrical distribution. A negative sign indicates that the tail of the distribution is skewed to the smaller values. If the sign is positive, as in this case, the distribution will be skewed to the right. Thus, the mean is located to the right of the mode 63 and the median. To help understand the implications of differentials in value of kurtosis, we may perceive the distribution of estimated direct coefficients consists of three areas: (1) area around its mean, (2) area between area 1 and its minimum, and (3) area between area 1 and its maximum. In this case, the shape of the distribution of the estimated direct coeffiCients is more peaked than the distribution of the "true" direct coefficients. This means that while the Simple Location Quotient technique overestimates the value of the direct coefficients around its mean, this technique underestimates the direct coefficients in area 2 and area 3. The Purghase Only Location Quotient (POLQYiTechnique In essence, the Purchase Only Location Quotient is not so different from the Simple Location Quotient, thus it also inherits most of the drawbacks of the Simple Loca- tion Quotient technique. The difference is in summation of total gross output of both the region and the nation. In- stead of using total gross output of the region and the nation in computing the location quotient, this technique uses only the total output of the industries which purchase from industry 1. The formulas are not presented here, be- cause they have been presented in Chapter III. This technique generates either a larger or smaller estimated direct coefficient than the Simple Location Quo- tient depending upon the size of the output of the 64 industries which are exluded, both at regional and national levels. Theoretically there are at least two advantages of this method over the Simple Location Quotient approach. They are: (1) By this procedure, the adjusted coefficient is not affected by large outputs of industries which are not directly related to the selling industries. (2) This technique can handle a situation where the needs of regional industries for output i, rel- ative to the needs of the national industries for output 1, are not the same as the ratio of total regional to total national output, a situation which the Simple Location Quotient technique can not handle adequately. The Purchase Only Location Quotient for each industry in the Tri-county region has been computed, and presented in Table 3. The regional direct coefficients generated by this method are presented in Appendix A-l. The statistical characteristics of the direct coef- ficients produced by the Purchase Only Location Quotient have been computed. In Table 5, these characteristics are compared to the statistical characteristics of the direct coefficients obtained from the direct survey table. Compared to the Simple Location Quotient, estimates generated by the Purchase Only Location Quotient differ 65 Table 5. The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (POLQ) Techniques Characteristics 1 2 DSM POLQ Mean .019 .010 Variance .002 .001 Standard deviation .050 .027 Kurtosis 31.249 52.828 Skewness 5.191 6.443 1 DSM = Direct Survey Method 2POLQ = Purchase Only Location Quotient slightly in two major characteristics: the mean and the kurtosis. The mean of estimates generated by the Purchase Only Location Quotient is slightly larger, but its kurtosis is smaller. This situation can be explained by the value of the Purchase Only Location Quotient, relative to the Location Quotient (see Table 3). In comparison to the values of the Location Quotient, the values of the Purchase Only Location Quotient are similar in 9 sectors, slightly larger in 9 sectors, and more than unity in two sectors. Thus, direct coefficients for both methods are similar in 11 sec- tors, because when the quOtient is larger than unity, value of the estimated direct coefficients are identical to the 66 national coefficients. Thus, difference occurs only in 9 sectors. That is probably the reason the mean of the estimated direct coefficients generated by the Purchase Only Location Quotient are just slightly larger than the mean of those coefficients generated by the Simple Location Quotient. In contrast the kurtosis and skewness of direct coefficients generated by the Purchase Only Location Quo- tient are slightly smaller than those generated by the Simple Location Quotient. The foregoing discussion indicates that this tech- nique makes only marginal impacts upon the results, as compared to the Simple Location Quotient technique. The Cross Industpy Quotient (CIQ) Technique One of the potential drawbacks of the Location Quo- tient technique is that only the size of the selling in- dustry is taken into account, although theoretically the relative size of the purchasing industry may also be of crucial importance. An advantage of the Cross Industry Quotient technique is that it enables import proportions to vary within the rows, whereas the Simple Location Quo- tient technique establishes import needs in constant pro- portion to the appropriate rows. The Cross Industry Quo; tient compares the proportion of national output of selling industry 1 in the region to that of the purchasing indus- h try j. The ijt cell of the national direct coefficient matrix is then adjusted according to the corresponding 67 cross industry quotient, as has been discussed in Chapter III. The cross industry quotient for each cell of the transaction matrix has been computed in this study and then used to estimate the direct coefficient matrix for the Tri- county region. The direct coefficient for the Tri-county region generated by this method is presented in Appendix A-l. The summary of the statistical characteristics of these estimates is presented in Table 6. Table 6. The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estima- ted Direct Coefficients (CIQ) Techniques Characteristics 1 2 DSM CIQ Mean .019 .022 Variance .002 .003 Standard deviation .050 .054 Kurtosis 31.249 20.142 Skewness 5.191 4.309 1DSM = Direct Survey Method 2CIQ = Cross Industry Quotient As can be seen from Table 6, in contrast to the Simple Location Quotient and the Purchase Only Location Quotient, the Cross Industry Quotient overestimates the 68 direct coefficients in an average of about 25 percent. While the two location quotient techniques adjust the na- tional direct coefficients downward drastically, the Cross Industry Quotient tends to overestimate the regional direct coefficients. There are some reasons why the Cross Industry Quo- tient technique overestimates the regional direct coef- ficients: (1) In contrast to the location quotient techniques, the adjustment process in the Cross Industry procedure takes place within the rows. Hence, there is a tendency that the Cross Industry Quotient will take more national direct coef- ficients than the location quotient techniques. For example in this study the estimated direct coefficients generated by this technique are similar to the national direct coefficients in 222 cells. In contrast, the estimated direct coefficients produced by the Simple Location Quotient and the Purchase Only Location Quotient are similar to the national direct coefficient in only 40 cells each. If the average of the national direct coefficients is significantly larger than the average of the regional direct coefficients, like in this study, then there is a tendency of the Cross Industry Quotient to overestimate the regional direct coefficient. 69 (2) The Cross Industry Quotient has a very dis— tinct property, that is: CIQij = LQi/LQj Clearly, if i = j, it follows that LQi = LQj. Thus, for every cell where i = j (those diagonal cells) the value of CIQij is unity and the estimated direct coefficient for the cells is identical to the corresponding national dir- ect coefficients. In other words, the implicit assumption is made that every sector can obtain all its requirements of output from its own sector locally, regardless of the size of the sector. It is felt by the author that this is a somewhat misleading assumption to make, particularly for a relatively small sector. From the discussion it may be concluded that: (1) there is a tendency that the Cross Industry Quotient overestimates the regional direct coefficients; (2) the average of the estimated regional direct coefficients gen- erated by this technique is the closest to the average of the national direct coefficients as compared to those dir- ect coefficients generated by the other methods tested in this study. The Modified Supply-Demand Pool Technique The Modified Supply-Demand Pool technique relies on computing the commodity balance between regional output of good i and the local requirements of good i. The local 7O requirement of good i is estimated by using national direct coefficients. If the commodity balance is positive, it is assumed that regional supply is sufficient to satisfy re- gional demand and that the national direct coefficients may be used in row i of the regional trade coefficients matrix. When the commodity balance for row i is negative it is assumed that the regional demand is larger than regional supply and importation of good i is necessary. In this case, the national direct coefficients of row i are re- duced by the ratio of regional output to the regional re- quirement. The Tri—county regional direct coefficients ahve been estimated by using this technique and the results are pre- sented in Appendix Table A-1. Before comparing the relative performance of the Modified Supply-Demand Pool technique to performance of the other reducing techniques in estimating regional direct coefficients, it is appropriate to discuss briefly the statistical characteristics of estimates produced by this technique (Table 7). As can be seen from Table 7, the Modified Supply- Demand Pool technique is similar to the Location Quotient technique in underestimating the regional direct coeffici- ents. As in the Location Quotient technique, in this pro- cedure downward adjustments of national direct coefficients are carried out if the balance of commodities for that row is negative. The estimated value of direct coefficients 71 Table 7. The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (MSDP) Techniques Characteristics 1 2 DSM MSDP Mean .019 .022 Variance .002 .001 Standard deviation .050 .033 Kurtosis 31.249 29.545 Skewness 5.191 4.834 lDSM = Direct Survey Method 2MSDP = Modified Supply-Demand Pool is identical with the value of national coefficients if the balance of commodities is positive. Thus, the national direct coefficients would never be adjusted upward. The difference between the Modified Supply-Demand Pool tech- nique and the Location Quotient technique lies in the dif- ference of factors used to adjust the national direct co- efficients downward. In Location Quotient techniques the value of the location quotient is used, while in the Mod- ified Supply-Demand Pool technique, the ratio of regional output and regional requirement is used. Since the ratio of regional output and the regional requirement average larger than the average of the location quotients, the mean of the estimated direct coefficient produced by the 72 Modified Supply-Demand Pool technique is larger than the mean of the same coefficient generated by the Location Quotient technique. RAS Me thod As has been pointed out in Chapter II, the RAS method was developed for the purpose of projecting input- output tables, using a base period matrix and projection period row and column totals introduced as constraints. In this study the RAS method has been used to project the 1958 national input-output table to regional dimensions by being made to conform with regional constraints. This study involved the use of the following sources of data: (1) 1958 Tri-county region total gross output (x) (2) 1958 Tri-county region total intermediate output (u) (3) 1958 Tri-county region total intermediate input (V) (4) 1958 The United States input-output coeffici- ents (Aij) The regional data were obtained through direct survey. The logic of this technique is simple. The adjust- ment process forms an iterative procedure. Each stage of iteration generates a new estimated total intermediate out- put and total intermediate input. The ratio of regional total intermediate output to estimated total intermediate output is used as an adjustment factor for corresponding rows. Likewise, the ratio of regional total intermediate 73 input to estimated total intermediate input is used as an adjustment factor for associated columns. Thus, adjust- ment processes take place in two directions, row and column. At some stages of iterations, in this study after 12 iter- ations, the value of estimated total intermediate output equals the value of regional total intermediate output, and in parallel the value of estimated total intermediate input equals the value of regional total intermediate input. Thus, the adjustment factor for both rows and columns reaches unity and the iterative process ceases. This method has been used to estimate the Tri-county regional direct coefficients, and the results are presented in Appendix Table A-1. The direct coefficients estimated by the RAS method will be compared to the "true" direct coefficients computed from survey table in the second part of this chapter. At this stage it is appropriate to compare their statistical characteristics. This comparison is presented in Table 8. As can be seen from Table 8, the RAS method in aver- age underestimates the value of direct coefficients. How- ever its kurtosis, was almost two times the kurtosis for the "true" direct coefficients. This means the distribu- tion shape of direct coefficients produced by the RAS method is more peaked than the distribution of "true" direct co- efficients. To explain this situation, the frequency dis- tribution of "true" direct coefficients can be divided into three areas: (1) area around its mean, (2) area between 74 Table 8. The Statistical Characteristics of the "True" Direct Coefficients (Survey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (RAS) Techniques Characteristics 1 2 DSM RAS Mean .019 .015 Variance .002 .003 Standard deviation .050 .053 Kurtosis 31.249 51.936 Skewness 5.191 6.969 lDSM = Direct Survey Method 2RAS = RAS area 1 and its maximum, (3) area between area 1 and its minimum. Information from Table 8 suggests that while the RAS method underestimates the value of direct coefficients in area 2 and area 3, this method underestimates the value of direct coefficients in the area around its mean. Because the total frequency of distribution in area 2 and area 3 is larger than the total frequency in area 1, the RAS method on the average tends to underestimate the "true" regional direct coefficient. Schaffer-Chu Iterative Method (SCM) The first step in the Schaffer-Chu technique is to calculate inputs requirements for producing regional output 75 j by using national direct coefficients. Then, regional sales are distributed in accordance with the national sales distribution patterns. Next, requirements and sales are compared, and a pool of surplus available for reallo- cations are constructed for each row i. If Pooli is larger than Needsi, each buying industry receives all its require- ments of product i, and the excess is exported. On the other hand, in sectors where pooli is smaller than needsi, the pool is spread by an iterative process among indus— tries in accordance with their relative need, until pooli diminishes to zero. In this study, pooli diminishes to zero after two iterations and then the iteration process ceases. The estimated direct coefficients obtained by using this method are presented in Appendix Table A-1. The value of estimated direct coefficients will be compared to the "true" direct coefficients (obtained from the survey table) later in the second part of this chapter. At this point, however, it is appropriate to compare the statistical characteristics of the estimated direct coef- ficients to the statistical characteristics of the "true" direct coefficients. This comparison is presented in Table 9. As illustrated in Table 9, the Schaffer-Chu tech- nique on the average overestimates the value of direct coefficient by around 40 percent. This situation can be analyzed in relation to the kurtosis of both distributions. 76 Table 9. The Statistical Characteristics of the "True" Direct Coefficients (Sruvey Method) Compared to the Statistical Characteristics of the Estimated Direct Coefficients (SCM) Techniques Characteristics 1 2 DSM SCM Mean .019 .026 Variance .002 .003 Standard deviation .050 .056 Kurtosis 31.249 15.799 Skewness 5.191 6.513 lDSM = Direct Survey Method 2SCM = Schaffer—Chu Iterative Procedure The distribution of "true" direct coefficients is more peaked than the distribution of the estimated direct coef- ficients. Assuming that the distribution of the regional direct coefficients consists of three parts: (1) area around its mean, (2) area between area 1 and its maximum, and (3) area between area 1 and its minimum. The informa- tion from Table 9 suggests that while underestimating the direct coefficients around its mean, the Schaffer-Chu technique overestimates the regional direct coefficients in area 2 and area 3. Because the total frequency of dis- tribution in area 2 and area 3 is larger than the total frequency distribution in area 1, therefore the Schaffer- Chu technique on the average overestimates the value of 77 direct coefficients. The difference in skewness is not singificant. Thus, the differences in the mean of direct coefficients generated by these methods may be attributed to the differences in kurtosis. Comparisons of Relative Efficiency of the Six Reducing Techniques in Estimating Direct Coefficients In this section, relative efficiencies of the six techniques tested in this study are compared. Relative efficiency means the degree of closeness in value of direct coefficients generated by each technique to the "true" direct coefficients obtained from the direct survey (assum- ing zero error). The relative efficiency of each reducing technique has been judged by using three comparison tests. These are: (l) the mean absolute percentage deviation, (2) the mean of relative change, and (3) the mean of similarity index. The prOperties of the comparison tests have been discussed in Chapter III. However, to facilitate interpre- tations of the results of this study, it is appropriate to discuss elaborately the properties of the comparison tests used in this study. Throughout the discussion of this chapter aij is defined as the direct coefficient of the direct survey table, and bij as the direct coefficient of the estimated table. The first comparison technique employed in this study 78 is the mean absolute percentage deviation (dij). The dij can be defined as follows: [a.. = b..| d.. = 11 ll_ 13 a.. 13 Two advantages of this statistic are that it is easy to calculate and to understand. For example, a dij equal to .50 means that the deviation is equal to 50 per- cent of the "true" direct coefficient. However, this tech- nique is deficient in other properties. The most serious deficiencies of this technique are: (1) It can not handle a situation in which aij is equal to zero and bij is positive. In this ., d.. is J 1] infinite if aij is zero. Thus, in the situation, case, regardless of the value of bi where aij is zero and bij is also zero no credit is given to the relative efficiency of the re- ducing technique. Because a zero aij is not uncommon in this type of study, this technique is likely not ideal. (2) Deviations between small coefficients affect the results far more than deviations between large coefficients. In other words, this technique may give different weights to the same absolute percentage deviation. To make clear this deficiency, one might consider two cases: 79 Case No. l Let a.. = .100 l] b.. = .001 13 Thus, _ L.100 - .001] _ Case No. 2 Let a.. = .001 13 b.. = .100 13 Thus, _ .001 - .100 _ dij - .OOId’ — 99.000 If we compare case No. 1 and No. 2, it is clear that even if the absolute deviations are the same in both cases, that is .099, the dij of case No. 2 is one hundred times larger than dij of case No. 1. In interpreting the re— sults, the reader must be aware of this problem. The second comparison technique used in this study is the mean of relative change. This technique lacks some of the deficiencies inherent in the mean absolute percent— age deviation technique. The mean relative change (Ri.) 3 can be defined as follows: Advantages of this technique are: (1) It could handle a situation where aij is equal 80 to zero. (2) Deviations between small coefficients affect the results in the same degree as with between large coefficients. However, this technique is deficient in another re- spect, that is, it ranges in value from zero to two. This value is not only awkward but also confusing. The third technique, the mean of similarity index, is the most satisfactory comparison technique used in this study. The similarity index can be defined as fol- lows: Iaij_- b Sij=l'(a..+b 13 J__ ) i.) ij It should be recognized that the closer the value of Sij to unity, the better the relative efficiency of the par- ticular reducing technique. In general, this technique lacks the deficiencies inherent in the first two comparison tests (this technique can handle a situation where aij is equal to zero, and its value is not confusing because it ranges from zero to one, however it is somewhat difficult to compute. The direct coefficients produced by the six reducing techniques tested in this study have been judged by using the three comparison tests outlined earlier, and the re- sults are presented in Table 10. Based upon data in Table 10, the six reducing tech- niques have been ranked in accordance with their relative 81 Table 10. Comparisons of the Relative Efficiency of the Six Reducing Techniques Used in This Study Comparison Techniques Reducing Techniques The mean abso- The mean of The Mean of lute percentage relative similarity deviations change index SLQ 1.099 .987 .507 POLQ 1.126 .980 .510 CIQ 2.230 .940 .530 MSDP 2.798 .942 .529 RAS .9399 .694 .653 SCM 1.759 .857 .572 l SLQ = Simple Location Quotient; POLQ = Purchase Only Location Quotient; CIQ = Cross Industry Quotient; MSDP = Modified Supply-Demand Pool; RAS = RAS; SCM = Schaffer-Chu Iterative Procedure. efficiency, and the results are presented in Table 11. As can be seen from Table 10 and Table 11, it is apparent that the RAS method produces the best estimates when judged by any of the three comparison techniques. Indeed, this seems logical, given that the technique re- quires a certain amount of survey data. However, the de- gree of its superiority is not significant. For example, the mean absolute percentage deviation for the best non- survey method, that is the SimpleLocation technique, is only around 10 percent higher than that produced by the RAS method. Consistently, the RAS method produces a 82 Table 11. Evaluation of Estimated Tables: Reducing Techniques Ranked for Each Comparison Test Comparison Tests Rank The mean abso— The mean of The mean of lute percentage relative similarity deviations change index 1 RAS RAS RAS 2 SLQ SCM SCM 3 POLQ CIQ CIQ 4 SCM MSDP MSDP 5 CIQ POLQ POLQ 6 MSDP SLQ SLQ Note: SLQ = Simple Location Quotient; POLQ = Pur- chase Only Location Quotient; CIQ = Cross Industry Quo- tient; MSDP = Modified Supply-Demand Pool; RAS = RAS; SCM = Schaffer-Chu Iterative Procedure superior result when judged by the mean of relative change and the mean of similarity index. Among the purely nonsurvey reducing techniques, there are more variations in the results. The Simple Location Quotient and the Purchase Only Location quotient emerge as being superior to the other reducing techniques when judged by the mean absolute percentage deviation test, however, both of the techniques perform poorly when judged by the mean relative change and the mean of similarity index. In contrast, the Schaffer-Chu and the Cross Indus- try Quotient techniques emerge as being inferior to the 83 other techniques when judged by the mean absolute percent- age deviation test, but they perform better when judged by the mean similarity index and the mean of relative change tests. These inconsistencies are probably due to the tend- ency of the Schaffer-Chu and Cross Industry Quotient to overestimate the smaller value of regional coefficients. This in turn will cause a significant increase in the value of the mean absolute percentage deviation for each over- estimated cell. On the other hand, when we overcome this problem, that is when we judged the Schaffer-Chu and the Cross Industry Quotient technique with the mean of rela- tive change and the similarity index, it seems that these two techniques perform better than the Simple Location Quotient and the Purchase Only Location Quotient techniques. The Modified Supply-Demand Pool technique, is the poorest reducing technique when judged by the mean abso- lute percentage deviation test. However, it is interest- ing that it emerges as being superior to the Simple Loca- tion Quotient and the Purchase Only Location Quotient techniques when judged by the last two comparison tests. As is apparent from Table 10, the relative effici- encies of the six reducing techniques appraised in this study are not really satisfactory to the author. Conse- quently it is appropriate at this stage to discuss briefly possible source of errors in estimating direct coeffici- ents, when using the six reducing techniques. 84 As has been pointed out earlier, the models in this study are based on the 1958, 78 by 78 input-output table of the United States and the 1958, 24 by 24 input-output table of the Tri-county region. Both matrices have been aggregated into a 20 by 20 standard matrix, which was used in this study. Probably by increasing the degrees of aggregation, the likelihood of including more firms with different production functions into one sector in- creased. Indeed this situation may introduce more aggre- gation error into the results of this study. For further study in this field, this writer recommends using a less aggregated matrix. Next, a more detailed examination of errors by sec- tor will be made. A close observation of the estimated direct coefficients generated by each technique reveals that the greatest deviations occur in sectors in which the regional economy is specialized. All reduction techniques appraised in this study performed poorly in estimating the direct coefficients for this type of sector. In- cluded in this type of sector are: (1) Motor Vehicles, (2) Medical, Educational Services and Nonprofit Organiza- tions, and (3) Chemical and Allied Products. Probably the sectors in which the regional economy is highly specialized have a different technology from the national average, and their input-output coefficients cannot be estimated by the shortcut techniques used in this study. The second group of sectors in which large deviations 85 are to occur are those in primary activities. In this group are included: (1) Other Agricultural, Fishery Pro- ducts and Services, and (2) Mining and Others. Agricul- tural crops in the Tri-county region, for example would include different activities from the United States av- erage. This may be even more true of mining. Many types of mining activities included in the United States sector do not exist in the Tri-county region. The last factor which may be responsible in intro- ducing errors into the results of this study is the size of the Tri-county region relative to the size of the United States. The Tri-county region constitutes only a small part of the United States. It seems that the smaller the region the less the probability that the economic structure of the region will resemble the economic struc- ture of the nation. And ceteris paribus the less similar the economic structure of a region compared to the na- tional economy, the more errors will be introduced in es- timating the input-output coefficients of the region from the national input-output table. Income Multiplier Analysis In this part the relative efficiency of the six re- ducing techniques in estimating the Type I and the Type II income multipliers will be compared. In comparing the results, it is assumed that the Type I and Type II multi- pliers produced by the direct survey method are "true" 86 (assuming zero error). Thus, the relative efficiency of a reducing technique is a measure of closeness of the mul- tipliers estimated by the technique to the "true" multi- pliers computed from the survey table. As has been pointed out earlier, the Type I income multiplier is the ratio of direct and indirect income change to the direct income change resulting from the delivery of one dollar to final demand by a given sector. The Type II income multiplier is the ratio of direct, in- direct and induced to the direct income change, resulting from the delivery of one dollar to final demand by a given sector. In computing the Type I income multiplier, it is assumed that the local household consumption expenditures (direct sales to local household) as being exogenous to the model. The Type II income multiplier assumes local house- holds consumption expenditures as being endogenous to the model. This means that local households are treated as another industry in the system. Households rent property, sell labor, provide financing, etc., and their purchases of locally produced goods and services are considered to be parallel with those emanating from other local indus- tiral sectors. It should be recognized, however, that when households are treated asaaprocessing sector, house- hold inputs are assumed to vary preportionally with house- hold output. As is true of other processing sectors, a 87 "production function" is assumed for households. Because households are in fact final consumers, this amounts to the assumption of a linear homogenous consumption function. Indeed, this is a somewhat rigid assumption. However, due to very limited resources, there is no attempt here to relax the homogeneity assumption by developing, for example, linear nonhomogenous consumption functions using national consumption data. However, there are strong econ- omic arguments for treating local households as an indus- try. When output changes in response to a change in final demand (assuming constant population), by definition house- hold income increases. The increment of income in turn theoretically results in new household consumption expend- itures and new savings. New expenditures generate reper— cussions of their own on the local economy. The Type II income multiplier is used to estimate the magnitude of these repercussions on an industry to industry basis. The Type I Income Multiplier The Type I Income Multipliers, based upon the direct coefficients estimated by the six reducing techniques have been computed in this study. The results, along with the Type I multipliers computed from the direct survey input-output table are presented in Table 12. The more complete results of multiplier analysis (including the direct, indirect and induced income changes) for each re- ducing technique are presented in Appendix B (Tables 1—7). 88 mmwm.a mom>.H mowH.H momm.a mmoo.a mmmo.~ mmam.a mmmeosz om mmvo.a homo.a bmmo.a wnmo.a homo.a vomo.a oomo.a Bmmzozomoz ma mmmH.H Nva.H vooa.H omhm.a mowe.a Ammo.a ommH.H moH>mmmmmm ma mmmm.a mmmo.a Homm.a vmoo.a vaH.H oomH.H ommm.a moH>mmmmDm NH ooma.a mmmo.a voaa.a momH.H Hooa.a mono.a mmoH.H zcmmmmmm ma mwma.~ ommm.a mmHm.H NovH.N mmnv.a vmov.a mamm.a muH>mmmmmm ma hmmb.a Homo.a omom.a mH~>.H mmmm.a ommm.a mmam.H odmemomz ea mmmo.a moom.a mo>H.H mmmm.H NmmH.H mmma.a hmow.a zzoumzdme ma oomm.a ooam.a hmmm.H oomm.a momH.H vhma.a meN.H dmdoomqm NH oom>.H mmoH.H mmao.a movn.a nmna.a NNFH.H mmmm.a meommmDQmHz HH hmam.~ mmmm.m mmma.m ~va~.m omoa.m Haoa.m mmhm.~ UHmm>oz oa mmmm.a vmmm.a mmom.a hmwm.a mmom.~ ono~.H mhom.a HomqmszU.m oaam.a monH.H momm.a mmoo.a mamo.a boom.a mBQmEZOZmHz w mama.a smoo.H momm.m omma.m HmmH.H mmmH.H mmmm.a meommquzmu m nonm.a Hwam.m hmmv.a wmmm.H hmoa.a >mva.a Hmmm.H meommeszm o anom.a mma~.H ovmo.a onmo.a mmao.a mHOH.H mmmv.a meammoooz m momv.m omoo.m momm.a oawm.m ooov.H oH>~.H mmoo.~ meommmmHmmdeo N swoo.m ~amm.~ moav.a ommm.a mm>~.H NNmH.H vaso.~ ammuoem>q H Sum 93 mom: 0H0 Odom 04m Ema wEmz . OZ mosqflsnowe osmosowm muouoom zosum mane CH omumwe moooflc sauce mcflospmm xflm way an ooooooum Hofiamwuasz mEOocH H wows on» on ooHomEoo Aoonumz >o>usm uomuflav :oflmmm Sassooiflue on» mo uoflamfluasz mEoosH H mama one .NH dance 89 In the following tables an abbreviated sector name is used, in addition to the sector number, for conven- ience in reading the table. The full sector names and abbreviations are as follow: Sector Abbreviated Number Sector Name l. LVSTOCKPRD 2. OTAGFISHPRDTS 3. WOODPRDTS 4. PRINTPRDTS 5. CHMCLPRDTS 6. MISNONPRDTS 7. PRIMA 8. FABMEPRDTS 9. MACHINELECT 10. MOVEHIC 11. MISDURPRDTS 12. ELPOGASA 13. TRANSCOMM l4. WHORTRAD 15. FBRSERVICE 16. PERSERAM l7. BUSSERVICE 18. REPSERVICE l9. MDEDNONPRT Full Sector Name Livestock and Livestock Products Other Agricultural, Fishery Pro- ducts and Services Furniture, Lumber and Wood Products Printing, Publishing and Allied Products Chemical and Allied Products Miscellaneous Non-durable Products Primary Metals Fabricated Metal Products Machinery including Electrical Motor Vehicles Miscellaneous Durable Products Electric Power, Gas and Santiary Transportation and Communication Wholesale and Retail Trade Finance, Insurance, Real Estate and Retail Personal Service and Amusements Business Services Repair Services Medical, Educational Services and Non-Profit Organizations 9O 20. MINOTHER Mining and Others In the same manner, in the following tables, the reducing techniques, are known by their initials as set below: DSM = Direct Survey Method SLQ = Simple Location Quotient POLQ = Purchase Only Location Quotient CIQ = Cross Industry Quotient MSDP = Modified Supply-Demand Pool RAS = RAS SCM = Schaffer-Chu Iterative Procedure The Type I multipliers discussed here should be considered first approximations, particularly for an econ- omy as "open" as that of Tri-county region. The Type II multipliers which will be discussed later, are larger in every case, and are more accurate estimates of the income changes produced in the area by changes in final demand. The latter shows the effects of successive rounds of con- sumer spending (including induced effects), in addition to the direct and indirect effects of increases in sales to final demand by each of the processing sectors. As is apparent from Table 12, we could classify the six reducing techniques into three groups: (1) techniques which in every sector underestimate the Type 1 income multipliers for every sector, (2) techniques which in every sector overestimate the Type I multipliers for every sector, and (3) techniques which overestimate the Type I 91 multipliers in some sectors, while underestimating them in the other sectors. In a more detailed fashion the interpretation of Table 12, (l) (2) (3) This can be summarized as follows: The Simple Location Quotient and the Purchase Only Location Quotient techniques underesti— mate the Type I income multipliers in all sec- tors. The Cross Industry Quotient technique overes- timate the Type I income multipliers in 14 sectors and understimate them in 6 sectors. The Modified Supply-Demand Pool technique over- estimates the value of the Type I multipliers in 7 sectors, while underestimating them in 13 sectors. The RAS method overestimates the Type I multi— plier in 5 sectors, while underestimating them in 15 sectors. The Schaffer-Chu technuque overestimates the Type I multipliers in all 20 sectors. fact is not entirely surprising. The value of estimated Type I multipliers to some extent depends upon the estimated direct coefficients. As has been discussed earlier, both the Simple Location Quotient and the Pur- chase Only Location Quotient techniques tend to underes- timate the direct coefficients. From 20 sectors of the Tri-county region, only in two sectors are the location 92 quotients greater than unity. In these two sectors (40 cells), the estimated direct coefficients are identical to the national coefficients, while in the remaining 18 sectors, the national direct coefficients have been ad- justed downward. In contrast, the Cross Industry Quo- tient and the Schaffer-Chu procedures take the value of the national coefficients in 222 cells and 206 cells re- spectively. Because the national direct coefficient in general is larger than the regional direct coefficients, the estimated Type I multipliers generated by the Cross Industry Quotient and the Schaffer-Chu techniques in most cases are larger than those produced by the Simple Loca- tion Quotient and the Purchase Only Location Quotient techniques. This finding is consistent with the fact that the value of the Type I multiplier of a large region is usually larger than that of smaller regions for a partic- ular sector. From the discussion we may conclude that the more likely a reducing technique takes the value of the national direct coefficients, the greater the possi- bility that the technique overestimates the Type I mul- tipliers. The relative efficiency of each reducing technique in estimating the Type I income multipliers has been com- puted in this study. Three comparison tests: (1) the mean absolute percentage deviation, (2) the mean of rela- tive change, and (3) the mean of similarity index, have been used to judge the reliability of estimates generated 93 by each reducing technique. The results are presented in Table 13. Table 13. Comparison of the Relative Efficiency of the Six Reducing Techniques in Estimating the Type I Income Multiplier Comparison Tests Reducing Techniques The mean abso- The mean of The mean of lute percentage relative similarity deviation change index SLQ .207 .242 .879 POLQ .196 .227 .887 CIQ .152 .138 .931 MISP .215 .215 .893 RAS .150 .130 .935 SCM .154 .135 .932 Based on data in Table 13, the six reducing techniques have been ranked in accordance with their relative effici- ency, and the results are presented in Table 14. As may be expected, it is apparent that the RAS matbod generates the best estimates when evaluated by the ‘Unree comparison tests used in this study. Probably this is due to the fact that the RAS method used a certain Mount of direct survey data. However, the degree of its Superiority is not significant. For example, the mean ab- SOlute percentage deviation of the second best technique, 94 Table 14. Evaluation of Estimated Type I Income Multi- plier: Reducing Techniques Ranked for Each Comparison Test Comparison Tests Rank The Mean abso- The mean of The mean of lute percentage relative similarity deviation change index 1 RAS RAS RAS 2 CIQ SCM SCM 3 SCM CIQ CIQ 4 POLQ MSDP MSDP 5 SLQ POLQ POLQ 6 MSDP SLQ SLQ that is the Cross Industry Quotient, is .152, while the RAS method's mean absolute percentage deviation is .150. Among the nonsurvey techniques, the Cross Industry Quotient emerges as being superior to the Simple Location Quotient and the Purchase Only Location Quotient techniques when judged by the three comparison tests. The Schaffer- Chu Iterative procedure emerges as being the second best when judged by the last two comparison tests. The Modi- fied Supply-Deman Pool technique is the poorest technique when judged by the mean absolute percentage deviation test, however, this technique produces a better result when judged by the last two comparison tests. 95 One general conclusion can be suggested from this discussion: all of the techniques tested in this study perform better in estimating the Type I income multipliers than in estimating the direct coefficients. For example, the range of the mean absolute percentage deviations in estimating the Type I multipliers is between .150 and .215, the range of this index is between .9399 and 2.798 when- ever the six techniques are used to estimate the direct coefficients. Thus, it seems that the use of these tech- niques to estimate the Type I multipliers is really prom- ising. We next proceed to a detailed examination of errors by sector. Below is the list of sectors, in which the greatest deviations do occur when we use these techniques to estimate the Type I income multipliers. Simple Location Quotient and Purchase Only Location Quotiént l. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Services 3. Primary Metal 4. Finances, Insurance, Real Estate and Rental Cross Industry Quotient 1. Chemical and Allied Products 2. Primary Metals Modified Supplnyemand Pool l. Livestock and Livestock Products 96 2. Other Agricultural, Fishery Products and Services 3. Chemical and Allied Products 4. Miscellaneous Nondurable Products 5. Primary Metals 6. Transportation and Communication 7. Business Services 1. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Services Schaffer-Chu Iterative Procedure 1. Other Agricultural, Fishery Products and Services 2. Chemical and Allied Products In concluding, it is fair to say that most of the reducing techniques appraised in this study perform poorly when used to estimate the Type 1 income multipliers in Livestock and Livestock Products, and Other Agricultural, Fishery Products and Services. The Type II Income Multipliers The results of the Type II income multiplier anal- ysis of the Tri-county regional economy are summarized in Table 15. The more complete results of multiplier an- alysis which include the direct, indirect and induced in- come change are presented in Appendix B. 97 hamm.m oavm.~ mmam.a mooh.a mmmo.a Homo.a Nomm.m mmmeosz om omoo.~ mmam.m mmmm.a mnmm.~ cwmm.a onaw.a mmmm.~ amazozomo: ma Nomm.a momm.a cmom.a vmmm.H ocov.a vmmm.a mmmo.a moH>mmmmmm ma mooo.~ mmom.~ mocn.~ voom.m mmom.a cmmm.a mmmm.H muH>mmmmDm NH -ma.m mmmm.a vowm.a mmm~.~ mmam.a hooo.a momm.H zmmmmmm ma vomw.m hmmo.m mvmw.o hmcm.m mooo.v ommm.m Homm.m ommemozz ea cmmm.m Homo.m cmao.m mmma.m Homm.a ooom.a mchm.m Esoumz.m mmmo.m UHmm>oz 0H Hono.m mmnc.a mmvm.H Hwom.~ wmmv.a chav.a Homm.a BumqmzHU.H meamdaooz m macm.oa anm.m nmom.m oooc.m mmmw.v cmm~.v mmh>.h meommmmHmodeo m mamm.m mnmn.h covm.m Nmmm.m voma.m momm.~ Homh.m ommxuoem>q a 20m mmm mom: 0H0 OAOE qu Ema mEoz .oz mosqesnooe memosoom muouomm wpsum mace Ga pmumme wooded (some.mswusoom xflm may Sn owooooum umflamwuasz mEoocH HH came on» on omummEoo Acosuoz >w>usm poouflov sewmom >DGSOUIMHB.o:u mo Howamfluasz wEoocH HH meme one .mH canoe 98 Theoretically, the values of the Type II income mul- tiplier are a constant multiple of the values of the Type I income multiplier. The ratio of the two multipliers 1 can be expressed as: 1 l - {h+Hr(I-A) 1 Hc} R = the ratio of the Type II and the Type I multipliers = direct coefficients matrix intrahousehold consumption coefficient = row vector of household coefficients 31:25:15 II = column vector of household consumption expendi- ture coefficients Aside from the apparent novelty of this formula, there is a practical gain in that once the Type I multipliers are known, it is no longer necessary to construct an aug- mented matrix which includes the household sector and ob- tain its inverse in order to derive the Type II income mul- tipliers. However, the reader must be aware that the ratio of the Type II and the Type I income multipliers is constant only if the total final demands is equal to total payments when the model is closed, that is when the household sec- tor is moved into the processing sector. In a situation lRichardson, Harry W., 1972, 9p Cit., p. 43. 99 where the total final demand is not equal total payments, when the model is closed, as in the Tri—county input-out- put model,1 the ratio of the two multipliers may not be constant. That is the reason that the ratio of the Type II and the Type I multipliers computed from the survey table (see Appendix B, Table l) are not equal to a con- stant. As can be seen from Table 15, the Simple Location Quotient and the Purchase Only Location Quotient underes- timate the Type II multipliers in all sectors. In con- .trast, the Cross Industry quotient and the Schaffer-Chu techniques overestimate the Type II multipliers in all but one sector. The Supply-Demand Pool underestimates the multipliers in 17 sectors, while overestimating them in 3 sectors. The most balanced technique is the RAS. While it overestimates the multipliers in 8 sectors, it underestimates them in 12 sectors. As shown in Table 15, the values of the Type II income multiplier in 5 sectors: (1) Livestock and Live- stock Products, (2) Other Agricultural, Fishery Products and Services, (3) Motor Vehicles, (4) Wholesale and Re- tail Trade, and (5) Finances, Insurance, Real Estate and Rental, are relatively larger than in other sectors. Based on their properties we may classify these 5 sectors into two groups: (1) Sectors in which the Tri-county lO'Donnell, John L., 33 al., 1960, Op Cit., p. 256. 100 region is highly specialized, and (2) Sectors which have been obtained through high degree of aggregation. In the first group, it seems logical that in sectors where the region is specialized the indirect and induced income change are relatively larger than the direct income change. Because in computing the value of the Type II income mul- tiplier we divide the direct, indirect and induced income change by the direct income change, we may expect that the Type II multipliers in these sectors are relatively higher than in the other sectors. In the second group, we arti- ficially enlarge the sectors by aggregation. It is poss- ible that the larger the sector, the larger the indirect and induced income change relative to the direct income change. So that it is not surprising that the Type II multipliers of these sectors are relatively higher than of those smaller sectors. From our discussions we may conclude that the value of Type II income multipliers may be very high in two groups of sectors: (1) sectors in which the Tri-county region is specialized, and (2) large sectors, which have been obtained through a high degree of aggregation. The relative efficiency of the six reducing tech- niques in estimating the Type II multipliers have been computed in this study. Three comparison tests: (1) the mean absolute percentage deviation, (2) the mean of rela- tive change and (3) the mean of similarity index, have been used to evaluate the relative efficiency of each 101 technique, and the results are presented in Table 16. Table 16. Comparison of the Relative Efficiency of the Six Reducing Techniques in Estimating the Type II Income Multipliers Comparison Tests Reducing Techniques The mean abso- The mean of The mean of lute percentage relative similarity deviation change index SLQ .318 .438 .781 POLQ .308 .391 .804 CIQ .191 .177 .911 MIDP .284 .293 .854 RAS .159 .123 .938 SCM .199 .173 .914 Based upon data in Table 16, the six reducing techniques have been ranked according to their relative efficiency in estimating the Type II multipliers. The results are presented in Table 17. If we compare the efficiency of the six techniques in estimating the Tyep II income multipliers to their efficiency in estimating the Type II income multipliers, we may conclude that the relative efficiency of each tech- nique in estimating the latter is lower. Indeed, this situation is not surprising because when a technique is used to estimate the Type II income multipliers induced 102 Table 17. Evaluation of Estimated Type II Income Mul- tiplier: Reducing Techniques Ranked for Each Comparison Test Comparison Tests Rank The mean abso- The mean of The mean of lute percentage relative similarity deviation change index 1 RAS RAS RAS 2 CIQ SCM SCM 3 SCM CIQ CIQ 4 MSDP MSDP MSDP 5 POLQ POLQ POLQ 6 SLQ SLQ SLQ income change is estimated in addition to direct and in- direct income change. Thus, it is only natural that some errors are introduced when the technique to estimate the induced income change is used. The results in general are promising. The superi- ority of the RAS method is apparent. Its mean absolute percentage deviation is only 15.88 percent. However, this situation is not unexpected, because this method employs a certain amount of survey data. Moreover, even the purely nonsurvey technique produced satisfactory results. For example, the Cross Industry Quotient, the second best technique when judged by the mean absolute percentage 103 deviation test, produced estimates which on the average deviate from the "true" value by only 19 percent. The Simple Location Quotient and the Purchase Only Location Quotient techniques are among the poorest when judged by all comparison tests. The Schaffer-Chu Iterative pro- cedure gives superior results among the purely nonsurvey techniques, when judged by the mean of relative change and the mean of similarity index. Thus, by observing the results, we may come to the conclusion that the use of RAS,tflueCross Industry Quo- tient and the Schaffer-Chu Iterative techniques to esti- mate the Type II multipliers is promising. Next, we could proceed to a more detailed examina- tion of errors by sector. A close observation of the estimated Type II income multipliers produced by each technique would reveal that the greatest deviations oc- cur in the following sectors. Simple Location Quotient 1. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Ser- vices 3. Miscellaneous Nondurable Products 4. Primary Metals 5. Mining and Others The mean absolute percentage deviations of these five sectors is 49.76 percent, while the mean of total deviation is 31.79 percent. 104 Purchase Only_Location Quotient 1. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Ser- vices 3. Miscellaneous Nondurable Products 4. Mining and Others The mean absolute percentage deviation of these four sectors is 51.51 percent, while the mean of total devia- tion is 30.81 percent. Cross Industry Quotient 1. Chemical and Allied Products 2. Business Services The mean absolute percentage deviation of these two sectors is 53.54 percent, while the mean of total deviation is 19.09 percent. Modified Supply-Demand Pool 1. Livestock and Livestock Products 2. Chemical and Allied Products 3. Primary Metals 4. Mining and Others The mean absolute percentage deviation of these four sectors is 60.06 percent, while the mean of total deviation is 28.37 percent. 105 l. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Services The mean absolute percentage deviation of these two sectors is 62.25 percent, while the mean of total de- viation is 15.88 percent. Schaffer-Chu Iterative Procedure 1. Chemical and Allied Products 2. Miscellaneous Nondurable Products The mean absolute percentage deviation of these two sectors is 54.47 percent, while the mean of total de- viation is 19.90 percent. This analysis concludes the discussion of the re- sults of this study. The summary and conclusions from the results are presented in the next chapter. CHAPTER V SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS Introduction In the last two decades, the demand for input—output models at the small region level has been increasing, and is reflected to some extent by the number of studies which have been completed and are currently in progress. In general this trend has been allied with an increasing awareness of planning as a continuous and cyclical pro- cess. In a planning process input-output models could be used to elaborate the implications of alternative strat- egies, they can help in the evaluation of these alterna- tives and consequently also could be useful in formulat- ing and reformulating the planning objectives. A major obstacle to the widespread application of input-output analyses at the regional level is a short- age of requisite data. An implementation of an input- output model requires extensive data which few other mod- els require. However, few other models possess the com- prehensiveness of the input-output analysis. Unfortunately data are rarely collected and published in a form directly appropriate for regional input-output analyses, so the 106 107 regional analysts must collect all or a part of the data through direct survey or can attempt to produce an input— output table from the available published data. As discussed in Chapter II, many attempts have been made to produce a matrix of regional direct coeffic- ient from a matrix of national direct coefficients. Yet efforts at such nonsurvey and minimum survey techniques have not been highly successful in the past. Thus a study on nonsurvey or minimum survey techniques could con- tribute greately to progress in empirical regional econ- omic analysis. Summary of this Study As has been pointed out earlier, an empirical input- output table constructed for the Lansing Tri-county region for the year 1958 by O'Donnell, epqgl., has provided this study a basis for testing of the relative efficiency of nonsurvey and minimum survey techniques, based upon the United States input-output tables for the year 1958. The purpose of this study is to integrate previous work on nonsurvey and minimum survey input-output techniques at the regional level by comparing results of the applica— tion of the most promising of the earlier nonsurvey and minimum survey tehcniques in a consistent way with an empirically derived survey-based input output model for the Lansing Tri-county region. The empirically derived input-output model of the Tri-county region is a regional 108 version of the traditional Leontief Open, static input-, output model, and thus the nonsurvey and minimum survey techniques tested in this study produced tables of this nature. The sectoring arrangement adopted in this study is presented in Table 2. The purpose of the aggregation pro- cedure was to reduce the size of a 87 by 87 national in- put-output table, and a 24 by 24 input-output table of the Tri-county region to a 20 by 20 standard matrix. The six reducing techniques appraised in this study divide naturally into three major categories. These are: a. Quotient Approaches Included in this category are: (l) The Simple Location Quotient, (2) The Purchase Only Loca- tion Quotient, and (3) The Cross Industry Quo- tient. b. Commodity Balance Approach Included in this category is the Modified Supply- Demand Pool technique. c. Iterative Approaches Included in this category are: (l) The RAS technique, and (2) The Schaffer-Chu Iterative Procedure. All the techniques operated on the national input- output matrix and attempt to adjust the coefficient to produce regional coefficients that adequately represents the Tri-county region's economic structure or interindustry 109 linkages. The basic assumption of all techniques is that the national technology is assumed the same with technol- ogy at the regional level and that the regional direct co- efficients differ from the national direct coefficients to the extent that goods and services are imported from or exported to other regions. This assumption clearly implies the constraint that the regional direct coeffici- ents must alwasy be less than or equal to the national coefficients. If the goods and services are exported to other regions, the direct coefficients of the export- ing industry are identical to national coefficients for the particular row, and if goods and services are im- ported from other regions, the national direct coeffici- ents are adjusted downward. Based upon the direct coefficients produce by each technique, the Type I and the Type II income multipliers were computed. An essential part of this study was to establish some objective means of evaluating the relative efficiency of each technique. The relative efficiency is a measure of distance or relative distance between the value of estimated direct coefficients, the Type I and the Type II income multipliers, and the "true" direct coefficients, the Type I and the Type II income multipliers computed from the survey-based input-output table. For this pur- pose three comparative tests have been employed. These are: (l) The mean absolute percentage deviation, (2) The 110 mean of relative change, and (3) The mean of similarity index. The mean absolute percentage deviation test has two major deficiencies. These are: (1) It can not han- dle a situation where the "true" direct coefficient is zero, and (2) In this test, deviations between small co- efficients affect the results far more than those between large coefficients. The mean of relative change and the mean of simil- arity index lack of these deficiencies. However, the mean of relative change has another problem, that is its value ranges from zero to two. This feature is not only confusing but also awkward. Thus, it may be suggested that the mean of similarity index is the most ideal of the three comparison tests employed in this study. Relative Efficiengy of the Six Reducing Techniques in Estimating Ehe Direct Coefficients When judged by the three comparison tests, it ap- pears that the RAS technique produces the best estimates. However, the degree of its superiority is not really sig- nificant. For example, the mean absolute percentage de- viation of estimates generated by the RAS method is only 10 percent lower than the mean absolute percentage devia- tion of estimates produced by the second best technique (the Simple Location Quotient). Among the purely non- survey reducing techniques, there is more variation in the results. The Simple Location Quotient and the Purchase 111 Only Location Quotient emerge as being superior to the other reducing techniques when judged by the mean abso- lute percentage deviation test. However, both techniques perform poorly when judged by the mean of relative change and the mean similarity index. In contrast, when judged by the mean of similarity index and the mean of relative change, the Cross Industry Quotient and the Schaffer-Chu Iterative techniques produced better estimates than those produced by the other techniques. These inconsistencies are probably due to the fact that the Cross Industry Quo- tient and the Schaffer-Chu techniques tend to overestimate the small regional direct coefficients. The Modified Supply-Demand Pool technique produced the poorest estimates when judged by the mean absolute percentage deviation test. However, it is interesting that it emerges as being superior to the Simple Location Quotient and the Purchase Only Location Quotient when judged by the other two comparison tests. Relative Efficiency of the Six Reducing Techniques in Estimating the Type I Multipiier It was observed that all of the techniques tested in this study perform better in estimating the Type I multi— pliers than in estimating the regional direct coefficients. For example, the range of the mean of absolute percentage deviation in estimating the Type I multipliers is between .150 and .215, while this range is between .939 and 2.798 whenever the six techniques are used to estimate the direct 112 coefficients. As may be expected, it was apparent that RAS method generated the best estimates of the Type I income multi- pliers when judged by any of the comparison tests. Among the purely nonsurvey techniques, the Cross Industry Quotient emerges as being superior to the Simple Location Quotient and the Purchase Only Location Quotient when evaluated by the three comparison tests. The Schaf- fer-Chu technique produced the second best estimates when judged by the mean of relative change and the mean of similarity index. The Modified Supply-Demand Pool technique produced the poorest estimates when judged by the mean absolute percentage deviation. However this technique emerges as being superior to the Simple Location Quotient when evaluated by the other two comparison tests. Relative Efficiencyyof the Sic Reducipg Techniques in Estimating the Type II Income Mdltiplier If the efficiency of the six methods in estimating the Type II income multipliers to their efficiency in estimating the Type II income multipliers is compared, it can be concluded that, in general, their efficiency in estimating the latter is decreased. However, this situa- tion is not really unexpected, because when a technique is used to estimate the Type II multipliers, in addition to estimating the direct and indirect income change, it also estimates the induced income change. Thus, probably additional error was introduced when the technique is 113 used to estimate the Type II income multipliers. The results of the Type II multiplier analysis were promising in general. The superiority of the RAS method was apparent. Its mean absolute percentage deviation was only 15.88 percent. Moreover, even the purely nonsurvey technique produced satisfactory results. For example, the Cross Industry Quotient technique, the second best technique when evaluated by the mean absolute percentage deviation test, produced estimates which on average de- viate from the "true" value by only 19 percent. The Simple Location Quotient and the Purchase Only Location Quotient techniques are among the poorest when judged by all comparison tests. The Schaffer-Chu Iterative pro- cedure gave a superior result when evaluated by the mean of relative change and the mean of similarity index. The Possible Sources of Estimation Errors In estimating regional direct coefficients, Type I and Type II income multipliers, the greatest deviations have occurred in three groups of sectors. The first group of sectors in which large deviations were expected to occur were those in which the Tri-county regional economy is highly specialized. In this study, this group was represented by the Motor Vehicles sector. It seems that sectors in which the Tri-county regional economy is highly specialized have different production functions than the United States average. Consequently, 114 their direct coefficients cannot be estimated from na- tional coefficients by using mechanical techniques. The second group of sectors, which we may call pri- mary activities, were represented in this study by three sectors: (1) Livestock and Livestock Products, (2) Other Agricultural, Fishery Products and Services and (3) Min- ing and Others. Probably it may be safe to assume that these high levels of errors can be largely explained by the difference of mix of economic activities within a given sector. For example, agricultural crops or fish products in the Tri-county region may include different activities from the national average. The last group of sectors in which large deviations were expected to occur were those sectors which have been obtained through a high degree of aggregations. By in- creasing the degree of aggregation, it is likely this will simultaneously increase the possibility of including more and more firms with different production functions into one sector. This situation may introduce more aggrega- tion errors into the results of the study. Conclusions and Recommendations In estimating the regional direct coefficients, none of the reducing techniques appraised in this study were entirely satisfactory to the author. The distance between the survey and nonsurvey direct coefficients matrices is still too large in absolute terms to be 115 acceptable. The Simple Location Quotient and the Pur- chase Only Location Quotient appeared to adjust some co- efficients downward drastically. In contrast, the Schaffer-Chu iterative procedure and the Cross Industry Quotient technique appeared to overestimate some coeffic- ients. This study revealed that the more frequently a re- ducing technique equates the regional direct coefficints to the national direct coefficients, the more likely that the technique overestimates the regional direct coef- ficients. While the RAS method performs slightly better than nonsurvey techniques, it must be remembered that the method employed a certain amount of survey data. Of course, data collection requires costs. It should be noted that (l) the RAS method produced estimates which only slightly better than the other techniques,(2) the RAS method likely is less cost effective than the other techniques. Therefore, from a cost point of view, the RAS method is not neces- sarily more effecient than the other techniques. Among the purely nonseuvey techniques, it was ob- served that the Simple Location Quotient and the Purchase Only Location Quotient produced superior estimates of dir- ect coefficients when evaluated by the mean absolute per- centage deviation test. However, when judged by the mean of relative change and the mean of similarity index, it appeared that the Cross Industry Quotient and the 116 Schaffer-Chu iterative procedure emerged as being super— ior to the other nonsurvey techniques. These inconsisten- cies were probably due to the tendency of the Cross In- dustry Quotient and the Schaffer-Chu techniques to over- estimate the small regional direct coefficients. This study revealed that the relative efficiencies of the reducing techniques were better when used to esti- mate the Type I and Type II income multipliers, than when used to estimate regional direct coefficients. It was observed that the RAS method (the best tech- nique when judged by all comparison tests) produced the estimates of the Type I and Type II multipliers with mean absolute percentage deviation of 15.0 percent and 15.9 percent respectively. Among the purely secondary data reducing techniques, the Schaffer-Chu procedure and the Cross Industry Quotient, consistently produced the best estimates of the Type I and the Type II income multipliers when judged by any of the comparison tests. Therefore, the most promising techniques for estimating the Type I and the Type II multipliers are: (l) the RAS technique, (2) the Schaffer-Chu iterative procedure, and (3) the Cross Industry Quotient technique. In estimating the regional direct coefficients, and Type I and Type II income multipliers it was found that greatest deviations occurred in three groups of sectors: (1) Sectors in which the regional economy is highly specialized, (2) The Primary Activity sectors, and 117 (3) Sectors which have been obtained through a high de- gree of aggregation. In order to obtain more acceptable results, it is recommended that a field survey be used to obtain regional direct coefficients for these three groups of sectors. It is also recommended that national and regional input-output statistics be published in greater detail (less aggregation), especially with regard to service in- dustries. This would be of considerable benefit to plan- ners and other regional analysts desiring to apply some kind of nonsurvey and minimum survey techniques. Further research and experimentation with the six reducing techniques appraised in this study is needed in order to improve their relative efficiency in estimat- ing direct coefficients, and the Type I and Type II in- come multipliers. If relative efficiencies of the reduc- ing techniques can be improved significantly, it will help planners and other regional analysts in making appro- priate response to policy questions by using input-output models. APPENDICES APPENDIX A ESTIMATED DIRECT COEFFICIENTS GENERATED BY THE SIX REDUCING TECHNIQUES, COMPARED TO THE "TRUE" DIRECT COEFFICIENTS COMPUTED FROM SURVEY DATA APPENDIX A ESTIMATED DIRECT COEFFICIENTS GENERATED BY THE SIX REDUCING TECHNIQUES, COMPARED TO THE "TRUE" DIRECT COEFFICIENTS COMPUTED FROM SURVEY DATA The Results of Direct Coefficients Estimation Direct requirements per dollar of gross output for each sector estimated by each reducing technique are presented in Table A-1. To facilitate comparison, these coefficients have been placed side by side with the "true" direct coefficients computed from survey data. In an ateempt to save the space, a sector appears in Table A-1 as its number, instead of its full name. Row numbers refer to producing industries, while the column numbers refer to purchasing industries. The sector num- bers and the full sector names are as follow: Sector Number Full Sector Name 1. Livestock and Livestock Products 2. Other Agricultural, Fishery Products and Services 3. Furniture, Lumber, and Woods Products 4. Printing, Publishing and Allied Industries 5. Chemical and Allied Products 118 Sector Number 6. 13. 14. 15. 16. 17. 18. 19. 20. 119 Full Sector Name Miscellaneous Non-durable Products Primary Metals Fabricated Metal Products Machinery including Electrical Motor Vehicles Miscellaneous Durable Products Electrical Power, gas and Sanitary Services Transportation and Communication Wholesale and Retail Trade Finances, Insurances, Real Estate and Rental Personal Services and Amusements Business Services Repair Services Medical, Educational Services and Non—Profit Organizations Mining and Others In addition each reducing technique appears in Table A-1 in its initials as shown below: SLQ : POLQ: CIQ : MSDP: Simple Location Quotient Purchase Only Location Quotient Cross Industry Quotient Modified Supply-Demand Pool RAS Schaffer-Chu Iterative Procedure Direct Survey Method 120 .o .o .o .o .o .o .o hH H hooo. mooo. voco. coco. mooo. Nooo. .o mH H COHo. .o mmoo. ohoo. coco. mmoo. .o mH H .o .o .o .o .o .o .o oH H Hooo. Hooo. .o Hooo. .o .o Nooo. MH H .o .o .o .o .o .o .0 NH H .o .o .o .o .o .o .o HH H .o .o .o .o .o .o .0 OH H .o .o .o .o .o .o .o m H .o .o .o .o .o .o .o w H .o .o .o .o .o .o .o h H mmoo. once. mmoo. vmoo. mHoo. NHoo. mmoo. m H Nooo. Nooo. Hooo. Nooo. Hooo. Hooo. omoo. m H .o .o .o .o .o .o .o v H .o .o .o .o .o .o .o m H cOHN. echo. OMMH. oon. Home. omno. OMHN. m H mmmH. ammo. mmmo. mmmH. mmbo. ommo. omvm. H H Sum mo>uomv unnecesmmoum mo HMHHOU Hoe mangwHHovmu uooHHp :onoH zu:DOOIHue one .HI< mHnme 121 oboe. mmoo. coco. Nmoo. mmoo. oNoo. HHoo. oH N omoo. mooo. mHoo. smoo. mHoo. HHoo. mmoo. NH N .o .o .o .o .o .o Nooo. NH N mNoo. mNoo. NHoo. mNoo. HHoo. mooo. mmoo. HH N .o .o .o .o .o .o .o OH N .o .o .o .o .o .o .o m N mooo. .o .o .o .o .o .o m N Nooo. mooo. Nooo. Nooo. Hooo. Hooo. .o n N oomo. HHoo. cmHo. oomo. mNHo. Nmoo. hooo. m N NoHo. mmoo. mnoo. NVHo. omoo. mooo. mmoo. m N .o .o .o .o .o .o Hooo. o N QONH. ooNo. ono. OONH. homo. vmmo. HmNo. m N NHHN. mHmm. HGOH. NNCN. homo. meo. mth. N N ommm. mth. HmmH. mHmm. ONGH. NwHH. hch. H N .o .o .o .o .o .o vmoo. ON H Nooo. mooo. Hooo. .o Hooo. Hooo. nNoo. mH H .o .o .o .o .o .o .9 NH H 20m mg mom: OHU OHOm OHm Ema CEDHOU 30m mmschnome AcmssHusoov .HI< mHome 122 omoo. mmHo. mNoo. mmoo. vNoo. mNoo. mCHo. HH m mooo. mmHo. Nooo. .o Nooo. Nooo. Nmoo. OH m mmoo. NNNo. «Hoe. mmoo. mHoo. mHoo. VOHo. m m mvoo. cmHo. vNoo. mvoo. mNoo. wNoo. mooo. m m mooo. mooo. mooo. mooo. mooo. mooo. mHoo. n m oooo. mooc. Hooo. mooo. Hooo. Hooo. HHoo. m m HNoo. mHoo. vHoo. HNoo. oooo. mooo. OHoo. m m mooo. Nooo. Hooo. mooo. Hooo. Hooo. Hooo. o m Nmmm. ammo. cmho. Nmmm. some. Nmmo. oomN. m m NHoo. mvoo. mooo. cHoo. mooo. mooo. MHoo. N m .o .o .o . .o .o .o Hooo. H m .o .o .o .o .o .o Hooo. oN N .o .o .o .o .o .o mono. mH N mmoo. OHNo. Hooo. «Hoe. omoo. vNoo. NOHo. mH N .o .o .o .o .o .o .0 pH N OHoo. OHoo. mooo. nooo. vooo. mooo. oooo. wH N ocHo. .o «moo. mmoo. mmoo. omoo. .o mH N Sum mdm mom: 0H0 OHOm OHw Ema :EDHOU 30m moqucnome .i 3 I. i \( Ill \ III I‘ 11 I'll .. o i I iii) . 1 II I. l:l"' 1‘14, .' Ill illii'in‘ 41.-.-.v' Ill 1 .I to II.. 0 I) ) -.o‘l,‘in‘i- Iliilillilli. i ! IoeoeHoeooc .Hue wanes 123 hooc. mmoo. HHoc. NHoo. mooo. mooo. QHoo. h o coco. Nooo. vooo. coco. Nooo. Nooo. Hooo. c o Hmoo. omoo. MNoo. Hmoo. HHoo. HHoo. coco. m o mmNH. mmco. mmmo. mmNH. Hmvo. hove. mmho. o o Nvoo. OHoo. Nmoo. Nooo. mHoo. mHoo. HNoo. m o mHoo. wmoo. HHoo. mHoo. mooo. mooo. hooo. N o Nooo. Hooo. Hooo. Nooo. Hooo. Hooc. .o H o mooc. Hooo. Hooo. mooo. Hooo. Hooo. ONoo. ON m .o voco. .o .o .o .o coco. mH m .o .o .o .o .o .o oooo. NH m .o .o .o .o .o .o HHoo. SH m mooo. wHoo. mooo. nooo. mooo. mooo. mooo. mH m Nooo. mHoo. Hooo. Nooo. Hooo. Hooo. wmoo. mH m OOHo. bHoo. vooo. oHoo. vooo. vooc. mooo. oH m Hooo. Nooo. Hooo. coco. Hooo. Nooo. Hooo. NH m Hooo. Nooo. .o .o .o .o coco. NH m Sum mdm mom: 0H0 OHOm OHm Ema :ESHOO 30m moqusnuoB HomscHusoov .Hld mHome 124 mOOO. HOOO. oOHO. hmOO. HNOO. ONOO. .O o m NNOO. HOOO. mOmO. NHNO. OvOO. hoOO. vOOO. m m ONOO. >OOO. OHNO. hmOO. hNOO. oNOO. NNNO. N m HHOO. .O NMOO. HHOO. oOOO. VOOO. .O H m hHOO. .O HOOO. NOOO. HOOO. HOOO. OHOO. ON o OoHO. memo. HHHO. ONOO. HmOO. omOO. hNNO. OH o NHOO. oOOO. OHOO. OOOO. mOOO. HOOO. OHOO. OH H OOON. HOON. «OOH. OOON. OOOO. OOOO. omov. 5H o >HOO. MNOO. MHOO. vHOO. OOOO. OOOO. ONOO. OH o mmOO. HnNO. ovOO. hmOO. OHOO. OHOO. OHHO. mH o nNOO. nNOO. HNOO. HNOO. mOOO. OOOO. mHOO. oH o OmOO. NOOO. ONOO. omOO. hHOO. hHOO. NOOO. mH o NOOO. oOOO. NOOO. NOOO. HOOO. HOOO. oOOO. NH o msoO. ammo. OOMO. movO. mOHO. NOHO. OONO. HH o mOOO. NOOO. oOOO. .O NOOO. NOOO. hHOO. OH v mOOO. omOO. AOOO. mOOO. NOOO. mOOO. HHOO. m o ONOO. moOO. ONOO. OHOO. mOOO. OOOO. hNOO. m o :0m max mow: OHU OHOm OHm Ema :E5Hou 30m moquccooe ..II 3.1!. 1.!1.|r iii 11'. '. l U!‘lltl 1.1.. -i‘l- ii I- 1..-! U‘ will“ 3 1 1|. .4). M . «11:11.91 I} ..Il.«‘ ..l .l D .13.. . 3. I I. ..(‘lllil'l Ill AcoscHucouv .Him oHQoB 125 ONOO. NOOO. OOOO. vOOO. HOOO. HOOO. OhHO. H O OHHO. .O OHHO. OHHO. OHOO. OHOO. HOOO. ON O NNOO. OHOO. OONO. HNOO. hmOO. OmOO. OHOO. OH O ONOO. HOOO. NOOO. NOOO. OOOO. OvOO. OOOO. OH O OOOO. HOOO. NOOO. HOOO. OOOO. OOOO. vOOO. NH O OOOO. OOOO. NHNO. OOOO. ONOO. ONOO. vOOO. OH O OOOO. NOOO. ONOO. OOOO. mOOO. OOOO. mOOO. OH O HHOO. .O OOOO. OHOO. OOOO. OOOO. HOOO. OH O NOOO. NOOO. HOOO. NOOO. HHOO. OHOO. OOOO. OH O OOOO. .O HHOO. mOOO. HOOO. HOOO. .O NH O OONO. HHOO. OOOO. MONO. NOOO. ONOO. OOvO. HH O vOOO. OOOO. HOOO. OOOO. OOOO. OOOO. NOOO. OH O NHHO. NHOO. MONO. HHHO. OmOO. ONOO. ONOO. O O OHOO. OOOO. mONO. OOOO. ONOO. ONOO. OHOO. O O OOOO. HHOO. OHNO. OOOO. ONOO. ONOO. .O O O OOOO. OOOO. OOOO. OOOO. OOOO. OOOO. HHOO. O O NOnN. omOO. NOhN. NOON. Ova. ommO. OOOO. O O 20m mg mom: OHU Odom OHm Ema CEDHOU 30m mwsochume HomosHucoov .HI< oHnme 126 NOOO. OOOO. NHOO. OOOO. OOOO. OOOO. OHHO. OH O HOOO. HHOO. OHOO. NNOO. OOOO. vOOO. OmOO. OH O ONOO. OOOO. OvOO. HOOO. ONOO. ONOO. OOHO. OH O OOOO. OOOO. OHOO. OOOO. OOOO. OOOO. ovOO. OH O OOOO. OHOO. OHOO. OHOO. OOOO. OOOO. OHOO. OH O OOOO. OOOO. vOOO. OHHO. ONOO. ONOO. oomO. OH O ONHO. OOOO. OHOO. OHOO. .OOOO. OOOO. OOOO. NH O HOHO. HOOO. ONOO. OOOO. NHOO. NHOO. NOHO. HH O OOHO. HOOO. ONOO. HOOO. OHOO. OHOO. OOHO. OH O OOOO. ONOO. OOOO. OHOO. mOOO. OOOO. OOOO. O O OHHO. ONOO. OHOO. OHOO. OOOO. .OOOO. OOOO. O O HOOO. OVOO. OHOO. OHOO. OOOO. OOOO. OOHO. n O Omom. Ova. OOvO. Omom. OOHO. OOHO. NOOO. O O NOOO. OOOO. NOOO. OONO. OmOO. OOOO. OOHO. O O ONOO. NOOO. OOOO. vOOO. NOOO. NOOO. NOOO. o O Omoo. OHOO. OOOO. ONHO. ONOO. ONOO. OOOO. O O OOHO. OOOO. ONOO. OOOO. OHOO. OHOO. OONO. N O Sum mg mom: OH O OHOm OHm Ema :EoHou 30m mosoHcsome HowocHucouv .HI< oHnme 127 HOOO. NOOO. HOOO. OOOO. NOOO. HOOO. .O OH O OOOO. .O HOOO. NOOO. HOOO. HOOO. .O OH O OOOO. OOOO. OOOO. OOOO. OHOO. NHOO. ONOO. OH O OOOO. OOOO. OOOO. OOOO. OHOO. OHOO. OOHO. NH O OOOO. OOOO. OOHO. OOOO. OOHO. NOHO. OOOO. HH O OOOO. OOHN. OONO. HOOO. OOvO. OOOO. OOOO. OH O OOOH. ONOO. HOOO. OOOH. OOOO. OOOO. OOOH. O O HQON. ONHH. HOOH. OOOO. OHVN. OOON. OOON. O O OOOO. ONOH. vOOO. OOOO. HOvH. OONH. vONN. O O OOOO. .O HOOO. OOOO. HOOO. HOOO. .O O O OOOO. ONOO. OOOO. NONO. OOOO. NOOO. OOOO. O O HOOO. HOOO. OOOO. NHOO. OOOO. OOOO. .O o O ovOO. OHOO. OOOO. OOOO. OOHO. OOHO. OOHO. O O OOOO. HOOO. HOOO. OOOO. NOOO. NOOO. .O N O .O .O .O .O . . HOOO. H O OOOO. HOOO. OHOO. OOOO. OOOO. OOOO. OHHO. ON O OOOO. OOOO. HHOO. OOOO. OOOO. OOOO. OOOO. OH O Sum mmmmmmm .OH OOOO.H NOON. OOOO. OOON.H OOOH. OOOO. OOOO. muH>mmmmDm .OH NOOO.H HOOO. OOOH.H OOOH.H ONOO. OOOO. OHOO. Edmmmmmm .OH OOO0.0 NOOO.N OOOH.O OHNO.H OOHO. OOOO.H OHOO. moH>mmwmmm .OH HOO0.0 OOOO.N OOO0.0 ONHO.H ONON. ONOO. OOOO. Odmamomz .OH OOOO.N OOOH.H OONO.H OOOO.H ONOO. OOOO. OOOO. zzoomzOS .OH HOOO.H OOOO. OOOO. OOOO.H OOON. OOOO. OOOO. BUMHmszOOZ .O OOOO.H OOON. OOOO. OOOO.H NHOH. HONO. OHOO. meommmzmmmmnmmmHmwq .H HOHHQ oeoosH OMWMWMW HOHHQ OHGO mEoosH OHGO Houomm (HuHoz omosocH . OOHH :H IHHHDS mEoosH uuouHcsH oeoucH OcHos OH HH OQOB .uomHHosH w .o H come uomuHocH .uoouHo MUOHHO . O m uoouHo . . mHnmu Oo>Hsm ECHO pwsHmubo .HOOOHV :oHOom OussouIHHB one Hem mHmHHmHuHDE OEOOGH .Hum meoe 147 NOOO.H OHNO. OONO. OONO.H OOHO. OOHO. OOOO. mmmeosz .ON OOHO.H NHOO. OONO.H OOOO.H OONO. OOOO. OOOO. BOOZOZOMOE .OH OOOO.H OOOH. ONNO. HOOO.H OONO. OOOO. OOOO. MUH>mmmmmm .OH OOON.H OHHH. OHOO. OOOH.H OOOO. OONO. OOOO. moH>mmmmDm .OH OOOO.H OOOO. OOOO. OOOO.H OOOO. NOOO. OHOO. Edmmmmmm .OH NHO0.0 HOOO.H OONO.H OOOO.H OOON. OONO. OHOO. MOH>mmmmmm .OH HON0.0 OOOO.H HOOO.N ONON.H OOOH. OOHO. OOOO. ommemomz .OH OOOO.H OOOO. OOOO.H OOOH.H OOOO. OOOO. OOOO. SEOOOZOOB .OH NOOO.H HOOO. .OOOO. OOOH.H OOOH. NOOO. NOOO. mmcwomHm .NH OOOO.H OOOH. HOOO. NNOH.H ONOO. OHOO. OOHO. OBQOOODQOHE .HH HOO0.0 OONO. OOHO. HHOH.N OOON. OOOO. OOOH. OHmm>oz .OH OOHO.H OOOH. OHOO. OOON.H OOOO. ONOO. OOOO. BOMHmszOOZ .O ONOO.H OOOH. NONO. OOON.H HOOH. OOOO. OHOO. mecmmmzmam .O OOOO.H OOON. HOOO. OOOO.H ONON. NOOO. OHOO. OZHOO .O NONN.N OOON. NONO. OHOO.H OOHO. HHON. OOON» weommzozmHz .O OONO.H OOOO. OHON. OOOH.H ONOO. NOON. OOON. OBOOOHOZOO .O OOON.H OOOH. OHOO. OOOH.H HOOO. OOOO. OOOO. OBQmOBszO .O NOON.H HOOH. HNHO. OHOH.H OHOO. OOOO. OOOO. meomdoooz .O ONON.O ONON.H OHON.H OHON.H OOOH. NOOO. OOOO. >mmmommmHmOH .H oEoonH uoHHo oEoucH cousccH HoHHm OHno oEounH OHnO Honoow IHuHsz cousan . ooHHOcH IHuHsz oEoonH uuoHHonH oeoonH Ooaooooum HH oQOB .uooHHocH w . H omwa uuoHHocH .uooHHc HooHHo . pooHHc ucoHuoso noHuoUOH onEHm onu On OoumEHumo .HOOOHO noHOom OD:SOOIHHB on» How mHoHHmHuHsE oeounH .Nim oHnoe 148 ONOO.H OOOO. OOOO. ONOO.H NOHO. OHNO. OOOO. mmmeozoz .ON OOOO.H OOOO. OHOO.H OOOO.H HONO. OOOO. OOOO. amazozomcz .OH OOOO.H OOOH. HONO. OOOO.H OONO. NOOO. OOOO. muH>mmmmmm .OH OOOO.H OOHH. OOOO. HOOH.H HOOO. OONO. OOOO. muH>mmmmDm .OH OOHO.H OHHO. HOOO. HOOH.H NOOO. OONO. OHOO. scammmmm .OH OOO0.0 OOOO.H OOOO.H ONOO.H OOON. OONO. OHOO. moH>mmmmmm .OH OOO0.0 OOOO.H OOOO.N OOON.H OOOH. OOHO. OOOO. ooz .OH OONO.H OOOH. OOOO. NOON.H OOOO. OOOO. OOOO. BUMHMZHOOmmmommmHmO¢eo .N OOOO.N OOOO. NOOO. OHON.H ONOO. OOOO. OHOO. OOOMUOBO>H .H oEoonH HoHHo oeoocH cousan HoHHm OHnO oeoonH OHnO Houoom IHHHoz cocoonH .uuoHHonH IHuHsz oeoonH uoouHonH oEoonH OdHusooum HH omOe .uooHHccH . H omwe uooHHonH .uooHHo uoouHQ . .uuoHHQ unoHuooO GOHHMOOH OHCO omonousm on» On oouoeHumo .HOOOHV :oHOom OHGSOOIHHB one Hem mHoHHmHuHSE oeounH .Olm oHnoB 149 OOOO.H OOON. OOOO. OOON.H OOHH. OONO. OOOO. mmmeozoz .ON OOOO.N HHOO.H OOOO.H OOOO.H OONO. MOOHO. OOOO. Emmzozomaz .OH OONO.H OOOO. NOHO. OOOH.H OOOO. OOHO. OOOO. moH>mmmmmm .OH OOOO.N OOOO. OOOO. ONOO.H NONN. OOOO. OOOO. muH>mmmmDm .OH OOON.N OOOO. OOON.H OOOH.H OHOO. OOOO. OHOO. 24mmmmmm .OH NOON.O OOOO.N OOO0.0 NOOH.N ONOO. OOON.H OHOO. moH>mmmmmm .OH OOO0.0 OOO0.0 OOO0.0 OHNO.H HOOO. OOHH.H OOOO. Odmsmomz .OH OOOH.O OOON.H OOOO.H OONO.H ONOO. OOOO. OOOO. SEOUOZOmB .OH ONOO.N OOOO. HOHO.H OOOO.H HOOO. OOOO. NOOO. oz .OH HOOO.N OOOO. OOOO. OOOO.H OONO. OOOO. OOOO. BUHHMZHOUOS .O OOON.N OOOO. OOOO. OOOO.H OOON. ONOO. OHOO. meommmzmmmmommmHmw¢eo .N NOO0.0 OOOO.H OHOO.H OOOO.H OOON. OOOO. OHOO. ommxuoem>H .H oEoosH eoHHm oEoonH cousan eoHHm OHCO oEounH OHnO Hoeoom (HeHoz coooccH . IHeHsz oEoonH eooeHonH oEounH euoeeocH mnHosooem HH omwe .eooeHosH . . H omOB eooeHonH .eooeHQ euoHHo eooeHo enoHeoso meemsonH mmoeu one On ooeoEHemo HOOOHO nOHOom OeCSOOIHee one HOH meoHHmHeHSE OEOOCH .Olm oHnoB 150 OOHO.H HONH. .ONOO. OOOH.H OOOO. ONOO. OOOO. mmmeosz .ON OOOO.H OOOO. NOOO.H ONOO.H OONO. OOHO. OOOO. emmzoznmoz .OH ONOO.H OHON. OOOO. OOOH.H OOOO. OOOO. OOOO. muH>mmmmmm .OH OOOO.N ONOO. ONNO.H HOOO.H OONO. OOOO. OOOO. muH>mmmmDm .OH OOOO.H OHOO. OOOO.H OOHH.H HOOO. OOOO. OHOO. Edmmmmmm .OH OOO0.0 OOOO.H HOOO.N OOHO.H OOOO. OHNO.H OHOO. muH>mmmmmm .OH OOO0.0 OHOO.N OHH0.0 OOOO.H OHON. OOOO. OOOO. Dememomz .OH ONHO.N NOOO. HOOH.H OOOH.H NOOO. HOOO. OOOO. Ezovmzema .OH OOOO.N OOOO. HOOH.H ONOO.H OOOH. OOOO. NOOO. émdwooqm .NH OOOO.H ONNO. OHOO. ONHO.H HOOH. ONOO. OOHO. meommmsomHz .HH OOON.O OHHO. OOOO. OONH.N OOHN. OOOO. OOOH. OHmm>oz .OH OOOO.H OOOO. OOOO. OOOO.H OOON. OOOO. OOOO. BomqmszU¢z .O OOOO.H NOOH. HOOO. OOOH.H HOOO. OOOO. OHOO. meommmzmflm .O OOOO.H OOOH. OOOO. OOOO.H OOOH. OOOO. OHOO. «szO .O OOOO.N ONOO. OOOO. OOOH.H OOOO. OOON. OOON. OBOOOZOZOHS .O OOOH.O ONOO. OOOO. OOOO.N NNOO. OOOO. OOON. mammmHUSnU .O OOOO.H OOOO. OOOO.H OOOO.H OOON. ONOO. OOOO. mBQmOeszd .O OOON.H HOHH. HONO. OOOO.H OOOO. OOOO. OOOO. meommoooz .O OOO0.0 OOOO.H HOOO.N OOOO.H OOOH. OOOO. OOOO. >mmmommmHmo¢eo .N OOO0.0 OOOO. OOOO.H OOHO.H OONH. OONO. OHOO. OOOMUOBO>H .H eoHHQ oEoonH OMWMMHW HoHHm OHCO oEoonH OHGO Hoeoom (HeHsz cousan .eooeHonH IHeHoS oEoocH eooHHonH oEoocH onHospoem HH omOB .eooeHonH . . H omaa eooeHonH .eooeHQ eooHHQ . eooeHo Hoom comeoolmHmmom coHHHooz one On ooeoEHemo .HOOOHO coHOom OeCDOOIHeB one HOH meoHHmHeHsE oEoonH .Oim oHnoB 151 OHOO.N NONO. OONO.H OOOO.H OOON. NNOO. OOOO. mmmeosz .ON OOHN.N OOOO. HHOO.H OONO.H OHNO. OOOO. OOOO. amazozomaz .OH OOOO.H ONOO. OOHO. NOOH.H HOOO. OOHO. OOOO. MUH>mmmmmm .OH OOON.H NNOO. ONOO. ONOO.H OOOO. OHOO. OOOO. moH>mmmmDm .OH OOOO.H OOOO. NOOH.H OOOO.H OHOO. OOOO. OHOO. aflmmmmmm .OH OOO0.0 NOOO.N OOOH.O OOOO.H OOOO. ONOO.H OHOO. moH>mmmmmm .OH OON0.0 OOHO.N OOON.O HOOO.H OOOO. OOOO. OOOO. o¢memom3 .OH HOOO.N OOOO. OOOO.H NOOO.H OOON. OOOO. OOOO. zzoomzOz .OH OOOO.H OOON. OONO. OOON.H OOOO. OOOO. OOOO. HOHHmszUmmmommmHm0¢eo .N OOO0.0 OOOO.N NHOO.N NOOO.N OOOO. OOOO. OHOO. OOOMOOBO>H .H oEoonH moHHm oEoonH oooomcH eoHHm OHGO oEoonH OHno Hoeoom i eHsz cocoonH _ (HeHsz oEoocH eooHHo:H oEoonH . .eooeHonH . . OnHooooem HH omOB .eooeHonH . . H omOB eooechH .eooeHo eooeHa eooeHa coneoz Odm one On ooeoEHemo .HOOOHO noHOom OensoonHeB one eoH mHoHHmHeHSE oEounH .Oim oHnoe 152 OHON.N HOOO. ONOO. OOOO.H NOON. OOOO. OOOO. mmmaosz .ON ONOO.N OHOH.H OOHO.H OOOO.H HOOO. OONO. OOOO. amazozommz .OH NOOO.H OOOO. OONO. OOOH.H OOOO. OOHO. OOOO. moH>mmmmmm .OH OOOO.N ONOO. ONOO. OONO.H OOOH. OOOO. OOOO. moH>mmmmDm .OH OOOH.N OOOO. OOON.H OOOH.H OOOO. OOOO. OHOO. Sfimmmmmm .OH ONO0.0 ONO0.0 OOO0.0 OOOH.N OOOO. HONN.H OHOO. mUH>mmmmmm .OH OON0.0 HNO0.0 OHO0.0 OONO.H OOOO. OONH.H OOOO. aemamomz .OH OONN.O OHON.H ONOO.H OONO.H OOOO. NOOO. OOOO. zzoumz¢ma .OH HOOO.N OOOO.H OOOO.H OOOO.H OONO. OOOO. NOOO. OmfiuomHm .NH OOON.N OHOO. OOOO. OOOO.H NOOO. OONO. OOHO. whommmoomHz .HH NOON.O OOOO. OOOO. OOHN.N OONN. OOHO. OOOH. UHmm>oz .OH HOOO.N OHOO. OONO. OOOO.H NONO. NHOO. OOOO. BUMHMZHOUOZ .O OOHO.N OOOO. OOHO. OOOO.H HOON. OOOO. OHOO. meammmzmmm .O OOOO.N OOOO. OHOH.H OOOO.N OOOO. HOOO. OHOO. dszm .O OON0.0 OHOO.H OOOO.H OOOO.N OONO. OHOO. OOON. meommzozmHz .O OOO0.0 OONO. OONO. OHOO.H OOON. OOOO. OOON. OBDOOHUSOU .O OOOO.H HHOO. OOOH.H OOOO.H HHOO. OOOO. OOOO. meommeszm .O NOOH.N HOOO. HOOO. HOOO.H OOON. OOOO. OOOO. menmmcooz .O NHON.OH ONO0.0 ONO0.0 NOOO.N OOOO. NOHO. OOOO. >mmmommmHmwH .H oEOonH HoHHm oeousH couponH HoHHQ OHno oEoosH OHnO Hoeoom IHeHoz ooODOGH uHeHsZ oeoocH eooeaonH oEoonH . .euoeHonH . . mnHosooem HH omOB .eooeHocH . . H oQOB eooernH .eooHHo eooeHQ . eooeHo oeooouoem o>HeoeoeH snuneomeonum one On ooeoEHemo .HOOOHV :OHOom headcouHeB one HOH mHoHHmHeHsfi oEoonH .Oim oHnma BIBLIOGRAPHY B IBL IOGRAPHY Borque, Philip J., Edward J. Chambers, J. S. Y. Chiu, et a1. 1967. Tpe Washington Economy: An Input-Output §Eua§. Business Studies No. 3. Seattle, Washington: Uniyer- sity of Washington, Graduate School of Business Admin- istration; Washington State Department of Commerce and Economic Development. , and M. Cox. 1970. An Inventory of Regional Input-Output Studies in the United States. Occasional Paper No. 22. Seattle, Washihgton: University of Wash- ington, Graduate School of Business Administration. Bureau of Business Research. 1965. A Preview of the Input- Output Study. New Mexico: University of’New Mexico, Albuquerque. Czamanski, S., and E. E. Malizia. 1969. Applicability and Limitations in the Use of National Input-Output Tables for Regional Studies. Papers and Preceedings, Regional Science Association. 23:65477. Isard, Walter. 1975. Introduction to Regional Science. Englewood Cliffs, New Jersey: Prentice:Hall, Inc. . 1951. Interregional and Regional Input-Output Analysis: A Model of Space Economy. Review of Econ- omics and Statistics 33:318-328. , and R. E. Kuenne. 1953. The Impact of Steel Upon the Greater New York-Philadelphia Urban Industrial Re- gion. Review of Economics and Statistics 35:289-301. Kokat, R. G. 1966. The Economig Component of a Regional Socioeconomic Moddl. IBM Technical Report 17-210. IBM, Inc. Advanced‘Systems Development Division. Lewis, Cris W. 1971. Regional Economic Development. The Role of Water. Logan, Utah: Utah State University Foundation. Malizia, E. E., and D. L. Bond. 1974. Empirical Tests of the RAS Method of Interindustry Coefficient Adjust- ment. Journal of Regional Science 14:65-78. 153 154 McMenamin, D. G., and J. E. Haring. 1974. An Appraisal of Nonsurvey Techniques for Estimating Regional Input- Output Models. Journal of Regional Sciences 14:191- 206. ' Miernyk, William H. 1965. The Elements of Input-Output Analysis. New York: Random House. Miller, R. E. 1957. The Impact of Aluminium Industry on the Pacific Northwest: A Regional Input-Output An- alysis. Review of Economics and Statistics 39:200-209. Moore, F. T., and J. W. Petersen. 1955. Regional Analysis: An Interindustry Model of Utah. Review of Economics and Statistics 37:363-383. Morrison, W. I., and P. Smith. 1974. Non-Survey Input- Output Techniques at the Small Area Level: An Eval- uation. Journal of Regional Science l4(l):l-14. O'Donnell, John L., 33 El- 1960. Economic and Population Base Study_of the Lansing Tri-County Area: An Inter- industry Relation Analysis. College of Business and Pubiic SerVice, E. Lansing, Michigan: Michigan State University, 319 pp. Peterson, R. S., and R. A. Wykstra. 1968. A Provisional Input-Output Study of Idaho's Economy. University of Washington Business Review, 27:11-27. Richardson, Harry W. 1972. Input-Output and Regional Economics. Throwbridge, Wiltshire: Redwood Press Limited, 294 pp. Schaffer, W. A., and K. Chu. 1969. Non-Survey Techniques for Constructing Regional Interindustry Models. Papers and Proceedings, Regional Science Association 23:83-101. Shen, T. Y. 1960. An Input-Output Table with Regional Weights. Papers and Proceedings, Regional Science Associationlézll33119l Stone, R., and A. Brown. 1962. A Computable Model of Economic Growth (A Programmeifor Growth, I). London: Chapman and Hall. , and A. Brown. 1965. Behavioral and Technical Change in Economic Models. In (E. A. G. Robinson, Ed.) Problems in Econgmic Development. New York: McMillan and Company LTD. pp. 434-435. 155 B. 1970. Regional Input-Output Analysis as a Tool for Natural Resources Management--The Colorado River Basin Economic Study. Unpublished paper delivered at aniinteragency work group meeting of participants in the Colorado Basin Study. U.S. Department of Agriculrure. Economic Research Service. 1970. Sector Income and Employment Multipliers: Their Interactions on the National Economy. Tech- nical Bulletin No. 1421. Washington D. C.: Government Printing Office. Department of Commerce. Bureau of Economic Analysis. 1964. The Interindustry Structure of the United States: A Report on the 1958 Input-Output Study. Survey of Current Business 44(11):lO-29. , Bureau of Economic Analysis. The Input—Output Structure of the United States Economy: 1967. Survey of Current Business 14:24-56. Yan, Chiou Shuang. 1969. Introduction to Input-Output Economics. New York: Holt, Rinehart and Winston. ‘I‘i. .til,