RESEARCH METHODOLOGY 4.1. Introduction
4.2. Quantitative method
4.2.2. Empirical framework
4.2.3.4. Research population
Burns and Grove (1993:776) define a research population as a unit that consists of ‘all the subjects’ a researcher wants to study. The population of this study includes households with social grant recipients in informal settlements from Freedom Park, Soweto, which covers a radius of 1.27 square kilometres (StatsSA, 2014d:1).
According to the Census data of 2011 (StatisticsSouth Africa), Freedom Park has a total population of 10 755 with 2 485 households, of which 1 164 are informal dwellings (StatsSA, 2014c:1). However, not all of the 1 164 households in informal dwellings qualify as part of the research population since not all of them have social grant recipients. Figure 4.1 below presents the geographical map of Freedom Park, Soweto.
64 Figure 4.1. Map of Freedom Park, Soweto
Source: StatsSA (2014)
4.2.3.4.1. The sample size and sampling method
A research sample is a fraction of the selected subjects from the total research population (Kotrlik & Higgins, 2001:43; Trochim & Donnelly, 2001:14; Yount, 2006:1).
In general, the sample size determination is based on the population size for the study, and Kotrlik and Higgins (2001:48) have suggested that studies with a total population that is above 1 000 but below 1 500 should have a sample size between 213 and 230 units. As noted earlier, Freedom Park has 1 164 households that reside in informal dwellings but the number of households that receive social grants in Freedom Park is unknown. The sample size of households for this study was 215, representing of the total number of households in Freedom Park.
The sampling method used was systematic random selection, which requires a fixed interval for the random selection of sample units (Elsayir, 2014:110; Rose, Grais, Coulombier & Ritter, 2006:290). In this study, every fifth household was systematically and purposefully included in the sample, and if the fifth household did not meet the research population selection criteria, the first adjacent household that meets the selection criteria was selected. The selection criteria required households to have at least one social grant recipient, a respondent that is willing to participate, is 16 years or older, and has in-depth knowledge of the household’s income, consumption and saving patterns. The criteria were set to ensure that relevant data on the savings behaviour of households with social grant recipients was collected and analysed using the different regression models discussed below.
65 4.2.4. Econometric techniques
The responses in the questionnaires was analysed as qualitative data with similar responses being grouped together and coded before analysis. Grouping or structuring and pre-coding qualitative data is considered an effective and efficient research analysis strategy in the social sciences (Jacelon & O’Dell, 2005:217-220; Russell, 1993:127; Onwuegbuzie, Dickinson, Leech & Zoran, 2009:5-6). The collected data was cleaned by the researcher before being captured using the University of Johannesburg’s Statistical Consultation Service (STATKON). The captured quantitative data in this study was analysed as cross-sectional data because the data was collected at one instance and not longitudinally (Olsen & St George, 2004:7).
The data was analysed using four regression models – tobit, normal OLS, robust OLS and probit regression models. All four were used in order to mititgate the weaknesses of individual models. An advantage of the normal and robust OLS regression models is that they measure the relationship between the dependent variable and independent variables (Pohlman & Leitner, 2003:124). Additionally, the robust OLS regression is used to avoid violation of OLS assumptions if the collected data have missing variables or outliers that might lead to the violation of OLS assumptions (Schumacker, Monahan
& Mount, 2002:10). In this study, there are outlying households that either save or consume more than others. This necessitates the use of the robust OLS regression model. A disadvantage of normal and robust OLS regression models in this study is that the sample has right hand truncation since it does not include rich people, and which means that it is not a full representation of the entire South African population of households with social grant recipients. Therefore, the nature of a right hand truncated sample, as in this study, requires the use of a tobit regression model.
Nakano and Zusman (2016:13) suggest that tobit regression models should be used when the sample is too small or truncated to fully represent the entire population. Using this model requires a researcher to have censored minimum and maximum values (Imai, Keele, Tingley &Yamamoto, 2015:8; Kohler & Rodgers, 1999:224). Therefore, the tobit regression measures the left-censors (minimum value) and right-censors (maximum value) of a dependent variable (Kohler & Rodgers, 1999:224&228; Vincent,
66 2010:1).For this study’s tobit regression model, a minimum value censored is R25 and R1 500 is the maximum value saved by the surveyed households.
Although the tobit regression model mitigates the weakness of the normal and robust OLS regression models, the tobit model in this study has a weakness as a result of the nature of data collection in this study. The primary data on savings behaviour of households with social grant recipients is collected from household financial handlers, therefore the researcher relied on the reporting of the household financial handlers some of whom saved while others did not. As a result, the researcher had to use a probit regression model since the sample has two groups, households that save and those that do not save. The two groups represent a binary variable based on the binominal responses of the dependent variable, that is a household saving or not (Hatfield, Kramer & Normand, 2015:9; Kohler & Rodgers, 1999:224&228). The binary variable values are that a household saves or a household does not save. Households that save have a value of one (1) and those that do not save have a value of zero (0).
The use of the probit regression model is used because the study sought to analyse the difference between households that save and those that do not.
4.3. Summary and conclusions
This chapter provides an outline of the quantitative research methods used in this study. A general model used for savings is discussed. Additional independent variables that influence poor household savings are discussed and presented in an equation. Data of the independent variables and the dependent variable (i.e.
household savings) is collected through the use of a questionnaire. The structured questionnaire was used to collect both quantitative and qualitative data of households with social grant recipients. The questionnaire had open- and close-ended questions with the latter focusing on numerical and statistical data. The former focused on information that cannot be easily quantified such as attitudes, experiences, opinions and abilities regarding savings. After data collection, this information was thematically coded into numbers, which were analysed quantitatively.
The questionnaire was guided by both thematic and empirical frameworks used in previous studies including Keynes’ (1936:20) basic savings equation. This is
67 complimented by a savings function with additional independent variables as has been done in several other studies. The additional variables are as follows: necessity, goods, consumption, normal goods consumption, luxury goods consumption, household size, social grant income, income pooling, age and gender of household financial handlers.
These were captured in the questionnaire administered to the research sample consisting of households with social grant recipients from the informal settlement of Freedom Park, Soweto. From the total population of 1 164 households in Freedom Park, 215 households were systematically sampled. Households that comprised of at least one social grant recipient formed part of the sample size of 215 and were asked to answer the structured face-to-face questionnaire.
Fieldworkers were hired and trained to ensure that the data collection processes was reliable and valid. The training entailed thorough understanding of the research topic, objective, questionnaire, research population and reduction of data collection biasness. The researcher and one fieldworker conducted a pilot study with 12 sample subjects and it was used to determine whether the sample population understood the questions and to address any possible validity and reliability issues. This led to the questionnaire being revised and validity problems were mitigated by ensuring that the questionnaire consisted of all factors that affect the savings behaviour of households with social grant recipients. Reliability problems were resolved by ensuring that there are no data collection biases. The data was captured with the University of Johannesburg’s Statistical Consultation Service (STATKON) and analysed using tobit, normal OLS, robust OLS and probit regression models.
Ethical clearance was obtained from the University of Johannesburg. The rights of the sample population to informed consent, privacy, anonymity and confidentiality prior to participating in the study was respected.
68 CHAPTER 5
ANALYSIS OF THE SAVINGS BEHAVIOUR OF HOUSEHOLDS WITH