Project topics and materials
Get Complete Project Material File(s) Now! »
Nature of variables
The dependant or outcome variable in this study was the preservation decision. It is a dichotomous variable as it was either a decision to preserve or not preserve accumulated funds. As per the conceptual model, the predictor variables or covariates are measures of bounded willpower, bounded rationality and rationality. Considering the discussion in the first part of this chapter, bounded willpower, bounded rationality and rationality were measured using a variety of methods which are discussed in more detail in the following sections.
Bounded willpower
The CFC scale provides two scores, one for the future subscale (seven questions) and one for the immediate subscale (seven questions). A five point scale was used resulting in scores for each subscale ranging from seven to 35. The scores for each individual are then averaged to determine a score ranging from one to five for each subscale. These average scores can be used in the form of interval data as the use of a full Likert scale, as opposed to individual Likert items, results in data that can be analysed as interval data (Carifio & Perla, 2007:115).
The BIS scale uses a four-point scale and there are 30 items leading to scores ranging from 30 to 120. In general studies using the BIS scale have three approaches to categorise the data. First, the approach used in most studies (see for example McLeish & Oxoby, 2007; Ottaviani & Vandone, 2011) is to use the index scores themselves in the form of interval data. Second individuals can be categorised as displaying high impulsivity (score 72 and above), normal levels of impulsivity (scores from 52 to 71) and low levels of impulsivity (scores below 52) (Stanford et al., 2009). This categorisation provides ordinal data. Last, a median split approach can be adopted (Potts et al., 2006) to categorise individuals as having high levels of impulsivity and low levels of impulsivity, which results in binary categorical data.
Bounded rationality
The financial literacy test has 13 items and therefore an individual score ranges from 0 to 13 providing interval data. In addition, the two subscales, basic financial literacy and sophisticated financial literacy, also provide scores out of five and eight respectively. From an analysis perspective, the aggregation of scores using factor analysis has been applied in some studies (Lusardi & Mitchell, 2009; van Rooij et al., 2012; Van Rooij et al., 2011), however, summing the scores on each section of the index and then standardising them has been found to preserve the relevant information for the analysis of financial literacy (Bateman et al., 2011:50). This finding ties in with other studies which have found that simple additive approaches to financial literacy indices produce very similar results to more sophisticated weighted approaches (Behrman, Mitchell, Soo & Bravo, 2010:11).
Therefore for the purposes of this study each individual will have a score for the basic, sophisticated and total financial literacy questions calculated by summing the total correct answers of the individual. The scores will then be standardised using sample means and standard deviations to create three indices. The standardised index scores provide interval data.
The self-reported level of financial knowledge as well as highest education level provides ordinal data. The question regarding whether advice was sought create nominal categories.
Rational variables
The variables collected to determine the influence of consumption smoothing and liquidity constraints on preservation decisions provide the following type of data:
• Age (interval);
• Salary and net asset value (bands = ordinal) and ordinal data from the self-reported financial need questions;
• Relationship status (nominal)
• Reason for moving job (nominal) and
• Use of funds (nominal).
Other variables collected
The other variable collected in the questionnaire was categorised as follows:
• Amount of funds available when moving jobs (ordinal data).
Assumptions and limitations of logistic regression
The logistic regression method of analysis is very flexible as it does not require normal distributions of the predictor variables, nor does it require linear relationships or equal variances (Tabachnick & Fidell, 2001:517). However there are a number of assumptions and limitations to this model which need to be assessed (Field, 2012:273).
First, predictors must not be highly correlated as this distorts results and leads to high standard errors. Multicollinearity in this study was assessed by reviewing standard errors and where necessary assessing collinearity statistics such as the tolerance and VIF values and eigenvalues (Field, 2012:297). These issues, where relevant, are assessed in Chapter 5 and 6. Second, sparseness of data needs to be assessed. In general as logistic regression makes use of Chi-squared distribution and tests, there should be no cells with frequencies less than one and no more than 20% of cells should have frequencies of less than five (Field, 2012:274). Therefore in this study cross tabs were used to assess cell frequency for all ordinal and nominal variables, based on this analysis certain categories were combined to eliminate low requency cells. More detail regarding the specific changes made is contained in Chapter 5.
Third, while logistic regression does not require that there is a linear relationship between ariables, it does require that all predictor variables that are entered as continuous variables (i.e. at an interval level) display a linear relationship with the log odds (or logit) of the dependant variable. If variables do not meet this requirement, they can still be used in the ogistic regression, however they must be entered as categorical variables and dummy variables are required for each level of the variable. The resultant logistic regression model becomes more complex and difficult to interpret and the tests of the impact of the predictor variable are less powerful as a result of the inclusion of multiple parameters (for the dummy variables) rather than a single parameter (Agresti, 1996:191).
CHAPTER 1 INTRODUCTION
1.1 OVERVIEW.
1.2 PROBLEM STATEMENT
1.3 RESEARCH OBJECTIVES .
1.4 THESIS STATEMENT.
1.5 DELIMITATIONS AND LIMITATIONS .
1.6 ASSUMPTIONS .
1.7 IMPORTANCE AND BENEFITS OF THE PROPOSED STUDY
1.8 RESEARCH DESIGN AND CHAPTER OVERVIEWS
CHAPTER 2 SAVINGS BEHAVIOUR
2.1 INTRODUCTION
2.2 DEVELOPMENT OF SAVINGS THEORY .
2.3 BEHAVIOURAL ISSUES IN SAVINGS DECISIONS
2.4 IMPORTANCE OF THE INDIVIDUAL AS DECISION MAKER
2.5 DEFINED BENEFIT AND DEFINED CONTRIBUTION RETIREMENT SCHEMES
2.6 THE INDIVIDUAL AS DECISION MAKER IN RETIREMENT PLANNING DECISIONS
2.7 CONCLUSION
CHAPTER 3 RETIREMENT FUND PRESERVATION – DEVELOPING A CONCEPTUAL MODEL
3.1 INTRODUCTION
3.2 EVIDENCE OF LACK OF PRESERVATION OF RETIREMENT FUNDS
3.3 POTENTIAL REASONS FOR EARLY WITHDRAWALS
3.4 POTENTIAL SOLUTIONS
3.5 PROPOSED CONCEPTUAL MODEL
3.6 PREVIOUS STUDIES .
3.7 CONCLUSION
CHAPTER 4 RESEARCH METHOD & RESEARCH INSTRUMENT DESIGN
4.1 INTRODUCTION .
4.2 DESCRIPTION OF OVERALL RESEARCH DESIGN.
4.3 METHOD
4.3.1 Research instrument
4.3.1.1 Predictor variables group one: bounded willpower
4.3.1.1.1 Time orientation self-report measures
4.3.1.1.2 Impulsivity self-report measures .
4.3.1.1.3 Summary of measures of bounded willpower .
4.3.1.2 Predictor variables group two: bounded rationality .
4.3.1.3 Predictor variables group three: rational factors
4.3.1.4 Outcome variable: preservation decision
4.3.1.5 Structure and order of questionnaire
4.3.1.6 Reliability and internal validity of questionnaire
4.3.2 Data collection .
4.3.3 Data analysis
4.4 LIMITATIONS .
4.5 ETHICAL PROCEDURES
4.6 CONCLUSION
CHAPTER 5 RESEARCH FINDINGS
CHAPTER 6 MODEL BUILDING AND HYPOTHESIS TESTING
CHAPTER 7 ANALYSIS OF FINDINGS – TOWARDS A SCIENTIFIC MODE
CHAPTER 8 CONCLUSION
