Pula: Botswana Journal of African Studies, Vol.14 No.1 (2000) Opinion polls in Botswana W. B. Molefe & B. S. Mguni Democracy Research Project University of Botswana Abstract Opinion polls are a relatively new phenomenon in Botswana but nonet~eless. a welco,!,e development as they inform both the candidates and the campaigners on pertment tssues facmg the electorate. Opinion polls becomes more visible and more discussed during every general election in Botswana. Unlike developed countries where poll taking begins more than a year before an election and continues until the last few hours of the election night, polls her~ m:ve been conducted at one time only during election years. Due to financial and logl~tlcal constraints they are not conducted at times when there are issues of public and natIOnal concern. Most of the polls so far polls have been conducted by the Democracy Research ~roject of the University of Botswana. However, as Bogart puts it 'the fundamental prmclple of statistical sampling, the basis for polling, seems hard for many people to understand '. Introduction One outstanding feature of our present society is its preoccupation with numbers. There ~s a desire to quantify every aspects of our every day lives. The communication medIa maintains a constant flow of such information. Confronted with this mass of numbers called statistics, the common man becomes bewildered and starts wondering how these numbers are arrived at and at what cost (Barnett, 19'88:1). The presentation of numbers is designed to keep us informed of the situation in the world around us. It is designed to place a discussion in the proper perspective as well as help government; business and the professionals make the right type of decision. th Public opinion polls are one of the greatest contributions to democracy in the 20 century. By asking a properly selected sample to express their views, wishes and interests, these unofficial surveys provide very useful information on what people are thinking ab~ut current issues. The most dramatic side of opinion pol\s is their ability to predict official elections. Although new in the Botswana context, opinion polls are a regular feature in many western industrialised countries, particularly at election time. The results of such polls may be advanced as a prediction of the outcome of a forthcoming presidential or general election. The central objective of opinion polls is to make an assessment of what peoples' attitudes or opinions are with respect to a particular issue. They can also be used as part of ~arket research by commercial houses as well as assess the popularity or otherwi~e. of dIfferent government policies and legislation (Datta and Mguni 1989:1). However, opmlOn polls .are not. easil~ ~ccepted, liked and approved by everyone. Much of the contro~ersy assocIated With opmlOn polls has to do with their use to predict the outcome of electIOns. The. two fundamental points of contention being the accuracy and reliability of results, and the Influe~ce of public opinion by results of opinion polls. ElectIon forecasts are the most publici sed of all the polling activities, but could be seen as. the least valuable in terms of what is the point of knowing approximately today :what.w.IlIbe known exactly tomorrow. To the politiCian, the interest is to know the trends In opmlOn. To the researcher conducting polls, the value of election forecasts is that they offer an opportunity of demonstrating the validity of the methods used. :ollowing the birth of ~rivate newspapers in the early 1980's, unscientific "s~aw polls were conducted occasIOnally by some of these papers. These essentially entaIled 122 asking questions to people met randomly at a shopping centre, railway station and other such places. By and large all opinion polls in Botswana have been conducted by the University of Botswana Democracy Research Project (DRP). The first such poll was conducted in 1989 and the last conducted in 1999. This article reviews the opinion polls conducted by the DRP in 1989, 1994 and the 1999. Methodology The methodology used for obtaining opinion poll results is the most crucial aspect as this determines the validity, reliability, relevance and accuracy of any such results. Just like other social surveys opinion polls are conducted using statistical survey methods. The sample design is decided upon the light of what is practically feasible as well as of what is theoretically desirable. There are 40 constituencies in Botswana, each of which has been divided into various polling districts. The polling districts have further been sub-divided into polling stations. Botswana is a large and geographically diverse country with sparse population in the western region, and dense population in the eastern region. This uneven population distribution has resulted in different levels of development of infrastructure, social services and job opportunities. The country is also not ethnically a homogeneous society. The Tswana group is numerically the largest and culturally dominant but there are significant minorities, which influence political behaviour in some regions and constituencies. For instance the Kalanga in both Francistown and the North East have in the past voted predominantly for the opposition parties. However, not all ethnic minorities support opposition parties. The cases of Ghanzi, Kweneng West and the Kgalagadi constituencies show overwhelming support for the ruling BDP. Some constituencies have over the years become politically marginalised, i.e. in an election any of the main contesting parties could win in such a constituency. The Kgatleng East and West constituencies are cases in point. Therefore one would want to take such factors into consideration so as to maximize the likelihood of the sample being representative of the country's political map (Molutsi and Tsie, 1989: 13- 15). For the opinion polls surveys conducted over the years by the DRP various aspects of the methodology are presented below. Opinion Poll- July, 1989 The opinion poll of 1989 was the Project's first to be conducted in this country. The purpose was really to introduce and test the applicability and acceptability of polls in the context of Botswana (Mguni, 1989). The fundamental question asked in this poll was - "Which candidate do you think would be a better representative in Parliament?" A multi stage sampling procedure was employed. A total of 953 eligible voters were interviewed. A purposive sample of seven constituencies was selected at the first stage. The sample consisted of the following: Kanye, Selibe-Phikwe, Gaborone South, Mochudi, Francistown, North-East and Molepolole The sample was selected so as to have rural and urban representation as well as to include some constituencies for which political observers expected the race to be either tight or little contest. For each constituency the number of registered voters for the 1984 general election was used to determine the required sample size. The sample size for a given constituency was worked out so that the estimation error for each desired proportion separately, was less than or equal to 8 percent. The second stage of the sampling involved the selection of polling districts withi~ a given constituency. In each constituency a purposive s~ple of a~leas~hal.fof the ~lhng districts was selected. A number of factors were taken Into consideratIOn In selectIng the sample of polling districts. These included, for example, geograph!c distrib~ti?n of the districts, voting patterns in the 1984 election, income group distribuuon and ehglble voter population densities according to the 1984 voter registration figures. 123 The third stage of the sampling involved the selection of polling stations within polling districts. For this stage, the same selection guidelines used to obtain the sample of polling districts were used. However, sampling of polling stations was often not necessary because some polling districts had one or two polling stations. " The fourth and final stage of the sampling involved the selection of dwelling umts around the selected polling station. There are generally no established geographical boundaries defining dwelling units for which the eligible voter can clearly be sai~ to belong to a given polling station. In order to select the sample of dwelling units for a gl~en polling station, it was necessary to establish a block of dwelling units for each pollIng station. This was achieved by using both natural and carefully selected boundaries. For each polling station block a working sampling frame was drawn up and used to select a random sample of the required number of dwelling units. Every eligible voter in a selected dwelling unit was interviewed. The number of required dwelling units selected in each polling station block was obtained using the assumption that there would be about 2.or 3 eligible voters per dwelling unit. The number of dwelling units covered was proportIOnal to its number of registered voters in 1984. Opinion PoD - June, 1994 This survey was conducted after the main political parties had already conducted their primary elections in most constituencies. The selection of constituencies was based on the following considerations: Rural/urban variations, the previously marginal constituencies, strongholds of each political party and the assessment of trends in new constituencies. The constituencies selected for the seven day long opinion polls were: Francistown West, Gaborone West, Kanye, Kgatleng West, Mahalapye, Molepolole, Selibe-Phikwe. A total of 3500 respondents were selected from the seven constituencies with the result subject to a 3 percent margin of error and 95 percent level of accuracy. The fundamental question that was asked in this poll was - "If elections were held today, which party would you vote for?" Opinion PolI- March, 1999 The sampling procedure was designed using a mixture of sampling schemes in multiple stages. Marginal constituencies were included in the sample and some constituencies were purposefully chosen to study the effect of the defections from the BNF leading to the formation of the BCP. Also party strongholds were chosen to understand the factors underlying that strength. The results of the survey were subject to a 3% margin of error and with a 95 percent level of confidence. The fundamental question for this poll as with the others was - "If elections were held today which party would you vote for?" A total of 4365 nd respondents from a total of 15 constituencies were interviewed during tl1 the week of 22 March to 27 March 1999. The constituencies covered were: Gaborone West, Gaborone South, Gaborone Central, Lobatse, Selibe-Phikwe, Francistown East, Francistown West, Maun/Chobe, Mogoditshane, Moshupa, Kanye, Shoshong, Serowe North, Kgatleng East and Ghanzi. The sampling procedure was designed using a mixture of purposive or judgmental sampling, cluster sampling and systematic sampling in multiple stages. First, purposive samplIng procedure was adopted in the selection of constituencies based on the results of ~he 1994 general elections. In Particular, "marginal" and "just safe" constituencies were Included In the sample. Marginal constituencies are those in which the winning party scored less than 50 percent of the votes and just safe constituencies are those in which the winning party scored a 51 percent to 60 percent votes. Second, some constituencies were purposefully chosen to study the effect of defections from the Botswana National Front (BNI? lea~ing to the formation of the Botswana Congress Party (BCP). Third, other constJtue~cles such as Serowe North, which are known to be strongholds of the Botswana Democratic Party (BDP) were chosen to understand the factors underlying that strength. 124 Fourth, cluster sampling was used in the selection of polling districts. Once this was done, a random sa~ple of polling stations, with probability proportional to size (PPS), was drawn. EssentIally, PPS involves assigning to each sample cluster a sequence of random numb~rs equal t? its size and then sampling systematically. In addition to these sampling techmque, a deliberate attempt was made to ensure that the sample was representative of the geographical distribution of the country as well as balancing the urban and rural components of the population. Based on the foregoing, the sample size and distribution were sufficient to make a national projection. The sample sizes are somewhat larger than required by the statistical method. But given the relative lack of familiarity with opinion polls by the general public and the unclear and sensitive issue surrounding the breakups of the largest opposition party at the time, it would be thought prudent to exercise maximum caution to avoid possible opinion poII errors. Specification of precision and sample size The statement of precision desired in an opinion poll could be made by giving the amount of error that a researcher is willing to tolerate in the sample estimates. The accuracy of a sample estimate refers to its closeness to the correct population value and since, normally speaking, it is not known, the actual accuracy of the sample estimate cannot usually be assessed, though its probable accuracy - which we term "precision" - can be assessed. It should however be noted that the accuracy of a statistical procedure could only be judged in the long run. Suppose the researcher seek the minimum sample size, n, that ensures that: Pr ~p - pI> d}~ a For some prescribed d and small a. The researcher needs to specify the tolerance d , and the risk a of not obtaining such tolerance. That is, the researcher wants the probability or chance of obtaining an estimate that is different from the true value in either direction by a specified amount, d, to be small. Inother words, the researcher's aim is to minimise the chance of underestimating or overestimating a proportion by a certain amount. A sample design is a plan by which the specified precision will supposedly just about be met at lowest cost. After the returns of the survey are in come the preparation of the estimates, and the appraisal of the precision of these estimates. The design of any sample is 2 predicated on the basis of certain assumptions regarding the variance a of the universe, or its coefficient of variation, r. Once the survey is finished comes the question, what is the precision of the results? It ought to be r/ ,J;;. , where n is the size of the sample, but were our advance calculations of the precision carriedout with the right valueof r ? And were the instructions regarding the sampling procedure followed faithfully? Did the mistakes and necessaryor wilfuldeparturesfromthe specifiedsamplingprocedureaffect r? These and a myriad of other questions are some of the issues that a researcher has to be content with in the conduct of a statistical inquiry. The number of registered voters in the last election in a particular constituency was used to determine the required sample size. Certainly this underestimate the total num~r of eligible voters who will vote in this election in that some people who were under votmg age are assumed to be of voting age. The DRP does not have a readily available pool of trained fieldworkers and therefore has to recruit and train these every time it has to conduct a survey. No matter how 125 intensive the training can be, there can never be substitute for experience as regards the selection of households and respondents, and the actual conduct of the interview. The importance attached to field workers following the selection procedures cannot be overemphasised. A deviation from the proposed sampling plan can have adverse implications on the results of the survey. The accuracy of a sample, and the risk of error that a researcher is willing to accept, are primarily determined by the sample size. In determining the sample size, standard parameters common for the type of categorical data generated by opinion poll questions is available. For example, for an estimated confidence level of 95 percent and a confidence interval of plus or minus 3 percent; the researcher can be sure that in 19 cases out of 20 a reported Proportion on any given variable will differ by no more than 3 percent in either direction from what would have been obtained by interviewing all eligible voters. Sampling theory allows that, within these parameters, reliable results can be generated with a minimum sample of 1,068 (Rea and Parker 1992:125-131). Results The results of opinion poll voter preference and the corresponding votes cast for the given constituency in the elections for the years 1989, 1994 and 1999 are presented in this section. In all these polls the statistical procedure used to collect the information has been demonstrably reliable and sound. Tables 1,2 and 3 show that the results of the 1989, 1994 and 1999 opinion polls are in agreement with the general election results though with cases of discrepancy in some constituencies. For instance the 1989 opinion polls predicted a victory for the Botswana People's Party in the North East constituency, but was won by the Botswana Democratic Party in the general elections. This discrepancy is however within the error limit of 8%. In 1994 the DRP polls predicted a win for the BDP in Selibe Phikwe constituency, but this was not to be as the constituency was taken by the BNF with a margin of about 18 percent. The error margin for this poll was 3 percent. The 1999 opinion polls and general election results show discrepancies for the Gaborone West, Gaborone South and Gaborone Central constituencies. The BDP was tipped to win the Gaborone West and Gaborone South constituencies that had in the past been strongholds of the BNF. This unexpected result could be explained by the split of BNF leading to the formation of the BCP _ some sup~orters of the BNF defected to the BCP, some remained in the party and the non- partIsan electorate was probably sceptical and hence the BDP had an edge over the BNF. In the Gaborone Central constituency, the BNF was favoured to win but the BDP emerged victorious. The effect of the split might have been downplayed by the opposition initially but became apparent in the general elections. The results of this poll were subject to a 95 percent confidence and error margin of 3 percent a e : ercenta2e V TbilP Constituency oter Preference and Vote Cast in 1989 Candidate/Party Gaborone South % Voter Preference % Vote K. Koma (BNF)- 40 G. MatIhabaDhiri (BOP) 42 Francistown I7 P. Balopi (BOP) 30 67 R. Sikwane (BNF) 30 7 M. Modise (BPp) 16 North East 9 K. Nkhwa (BPP)- 22 43 C. Butale (BOP) 33 39 P. Woto (BNF) 36 MOlepOlole 8 O. Kwelagobe (BOP) 6 39 Kanye W. Seboni (BNF) 55 II L. Tlhomelang (BFP) 15 36 A. Mogwe (BOP) 46 Mochudi 19 R. Molomo (BOP) 12 31 Selibe Phikwe J. Pilane (BNF) 31 13 K. Morake (BOP) 38 31 G. Saleshando (BNF) 38 32 37 126 Tbl2P a e : ercen a2e Vto er Pr~ e erence an dV ote Castml 994 Constituency Party % Voter Preference % Vote Francistown West BOP 38 34 BNF 28 40 BPP 16 27 Gaborone West BOP 28 28 BNF 48 72 BPP I 0 Kanve BOP 30 37 BNF 43 55 IFP 6 8 Kgatlenl( West BOP 41 40 BNF 48 60 Mahalapve BOP 59 62 BNF 25 36 Molepolole BOP 45 72 BNF 33 28 SeliOOPhikwe BOP 43 40 BNF 42 58 BPP 3 2 Table 3: Percentage Voter Preference and Vote Cast in 1999 Constituency Party % Voter Preference % Vote Gaborone West BOP 39 35 BCP 6 19 BNF 25 43 Gaborone South BOP 33 38 BCP 6 4 BNF 27 57 Gaborone Central BOP 27 48 BCP 5 16 BNF 32 31 Lobatse BOP 31 41 BCP 3 7 BNF 41 51 Selebi Phikwe BDP 41 48 BCP 15 31 BNF 14 20 Francistown East BDP 42 51 BCP 5 17 BNF 15 18 Francistown West BDP 42 49 7 19 BCP 16 19 BNF Maun/Chobe 60 54 BOP 9 15 BCP 8 12 BNF 49 51 Mogoditshane BDP 6 13 BCP 22 36 BNF 49 52 Moshuoa BOP 3 4 BCP 26 30 BNF 39 Kanye BDP 33 I 4 BCP 53 BNF 38 71 70 ShoshonI! BOP 2 4 BCP 26 BNF 16 81 92 Serowe North BOP 3 BCP 1 127 BNF 7 5 Kgatleng East BDP 36 40 BCP 9 31 BNF 36 29 Ghanzi BDP 44 52 BCP 9 12 BNF 24 36 Interpretation and Limitations of PoDs . There is considerable controversy surrounding the use of opinion polls results dUrIng an election campaign. Opinion polls are not easily accepted, liked and approved by every.one. Much of the controversy associated with opinion polls has to do with their use to predIct the outcome of elections (Mguni, 1989:3). It is felt in some quarters that polls influence and even create public sentiment. Many people find it difficult to believe and understand how one can predict the outcome of an election on the basis of information collected from only a few people. It should however, be said that results of polls should not be taken as predictions but merely as the popular sentiment at the time of the interview. opinion polls should not be taken as election results but merely as people's perceptions and viewsabout politicalviews and processesat a particular time. Nevertheless, they are important because they assist politicians to know who is ahead in the race at constituency, at .any particularpoint in time and issues that influence voting trends' (Democracy Research Project, 1999). The reason is that voting preferences can change drastically in the latter weeks of a campaign, making it extremely difficult to make any predictions. Also attitudes andopinions of people can change very rapidly with the passage of time and as a result of the occurrence of certain events or decisions which are political. . The DRP's opinion poll of March 1999 was held before most, if not all, political parties had conducted their primary elections. The general elections were held in October 161h, eight months later. Now if the political electorate in Botswana votes a candidate as opposed to a party, then one would certainly expect the outcome of the primary elections to influence voting preferences. It has been shown in some instances that in Botswana, the vote is for the party and not the candidate; the personal strength and capability of the candidate is not the only guiding factor. The Botswana Congress Party (BCP) is a breakaway party from the Botswana National Front (BNF). All the eleven BCP members of Parliament, who had defected with the seats they won under the BNF, lost in the 1999 general elections. This could be seen as indicative of the fact that they were voted into Parliament by virtue of being candidates of the BNF and not on the basis of their individual popularity. These candidates were "popular" because of the BNF. On the other hand the BNF had fielded "relatively weak" candidates and controversial characters, and yet won the elections. Besides 'predicting' election results, opinion polls can throw light on the influences that determine the direction of popular vote during the course of a campaign. The manner in which opinion crystallises can be verified on the basis of the facts collected. When the voters can be classified by demographic characteristics, economic status and other factors, polls could be used to identify different opinion alignments by group. In this manner polls can add materially to our store of political knowledge. From the standpoint of sampling it should be emphasised that the political public is n~t ~ homo~eneous population but is divided up into groups between which there are sIgmficant differences of perceptions and interests. Thus there are men and women living in diff~rent geographical areas and earning their living as farmers, manual workers, the profeSSIonals, and so on. Others are unemployed and do not have a means of survival except as dependants. It is clear that different occupations produce different levels of 128 income and different living standards. The interaction of these forces divides the public into groups and stimulates common attitude within the groups. Obviously people who live differently are expected to at least think differently. Demographic characteristics and economic status such as age, sex, religion, language, occupation, political affiliation, education and the general cultural background are the basic determinants of people's experiences and opinions. Conclusion Opinion polls should not be taken as election results but merely as people's perceptions and views about political views and processes at a particular time. Nevertheless, they are important because they assist politicians to know who is ahead in a race at a given constituency, at any particular point in time and issues that influence voting trends. Without opinion polls the much talked about and valued 'will of the people' in a democracy would be even more difficult to determine. The experience of Botswana is that opinion polls work and inform both the political process and the people. Polls have become a mechanism through which the public becomes sensitised to its own needs and self-conscious about its own collective stance. References Barnett, V. Elements of Sampling Theory. Edward Amold, London (1988) Bogart, L. Polls and the Awareness of Public Opinion, 2nd edition. John Wiley & Sons, New York (1972) Deming, W.E. Some Theory of Sampling Dover Publications, Inc., New York (1966). Democracy Research Project 'The General Elections Opinion Polls Results, March 1999' (Unpublished). Mguni, B.S., 'Report on the July 1989 Opinion Polls'. 1989 (Unpublished). Mguni, B.S. & Datta, K. 'Report on Exit polls of 1989 general elections in Botswana' 1989 (Unpublished). Molutsi, P. & Tsie, B. 'Political Mobilisation and Strategies for Elections: A Report of the Study of the 1989 Botswana Elections'. 1989 (Unpublished). Rea, L. M. & Parker, R.A. Designing and Conducting Survey Research. A Comprehensive, Guide. Jossey-Bass Publishers, San Francisco (1992) Som, R.K. A Manual of Sampling Techniques, Heinemann, London(l976) Yates, F. Sampling Methodsfor Censuses and Surveys, 3'd edition. Griffin, London (1971). 129