• No results found

Data analysis techniques

As outlined in the previous chapter, the questionnaire was designed to address four ROs. The questionnaire had five sections with 49 questions, including one open- ended question. Each section aligned with one RO. The open-ended section that sought to address the last RO will be discussed in the next chapter.

In this study, the data analysis was performed with the assistance of SPSS software version 20.0. The relationship between independent variables (e.g., gender, region of origin, length of stay in Tasmania) and dependent variables (e.g., migrants’ experiences or views on food security) were analysed (Faherty, 2008). The independent variables consisted of categorical or nominal data that had

independent response categories and the dependent variables consisted of ordinal data such as Likert-scale responses (Huizingh, 2007). The questionnaire included 15 independent and 33 dependent variables. The independent variables were factors that influenced the respondents’ experiences, perceptions and attitudes towards food security in Tasmania. The statistical analysis suggested using SPSS should enable the determination of the interconnections and relationships between the two types of variables.

Two types of statistical analysis were performed to analyse the questionnaire responses: descriptive and inferential statistics. All collected data is either

categorical or ordinal. Thus, descriptive statistics were conducted initially to provide information such as the frequency and proportion of responses within each

category of the different variables. Inferential statistical techniques were employed where possible to identify the relationship between the respondents’ demographic background, experiences, perceptions and attitudes towards food security in Tasmania as well as to determine the significance of the results.

The distribution of the data was examined using the skewness and kurtosis of the data thereby ensuring that the appropriate statistical analyses could be adopted. Skewness and kurtosis provide an indication of the symmetry and peakedness of the distribution of the data (Pallant, 2013). Skewness and kurtosis values were far away from 0 indicating that the distribution of collected data was not normally distributed (see Appendix 8). Therefore, the statistical techniques chosen to analyse the data were the non-parametric such as Chi-square (χ2) tests, followed by ordinal logistic regression analysis to identify any interconnections and relationships between the variables.

4.2.1 Chi-square (χ2

) tests

Chi-square (χ2) tests were employed to evaluate the dependence among variables and to examine whether the socio-demographic factors including gender, region of origin, length stay in Tasmania, marital status (independent variables) are related to each Likert-scale outcome variable such as the migrants’ experiences and views on their food security in Tasmania (dependent variables). It is important to remember with Chi-square (χ2) test, that the number of the cells in the contingency table with the expected number of counts of <5 must not be >20% of the total number of cells. When the number of cells with the expected number of counts of <5 exceeds 20%, the table needs to be reorganised by merging the columns or rows so that the expected number of counts of <5 is not >20% of the total number of cells. For instance, the set of choices ‘Strongly agree and agree’ or ‘Very important and important’ can be merged into one group, depending on the context of the test. An alternative test to use when Chi-square assumptions are violated (i.e., >20% of expected cells <5) is the Fisher’s exact test. The results of the Chi-square (χ2) tests performed are shown in table A, B and C (See Appendix 9).

Chi-square (χ2) tests only demonstrate an association, they do not provide a measure of the strength of the relationship between the variables as well as the nature of the differences between the groups (Munro, 2005; Pallant, 2013). Those factors that were shown to be significant (p values ≤0.05) underwent ordinal logistic regression.

4.2.2 Ordinal logistic regression

Ordinal logistic regression analyses were conducted to determine which

independent variables have a statistically significant relationship to the dependent variables (Laerd Statistics, 2014). In the test, four criteria must be met in order for the result to be valid: the dependent variable must be ordinal; one or more

independent variables can be ordinal or categorical data; no multicollinearity should exist within the data (multicollinearity occurs when two or more independent variables are highly correlated with one another) and; the proportional odds need to be determined (i.e., each independent variable has an identical effect at each

cumulative split of the ordinal dependent variable). The Odd Ratio (OR) was also calculated. OR is a relative measure of effect that allows comparison within the independent variables (whether they had higher or lower values compared to other groups). For example, the OR may show that male migrants are less likely to know about food in Tasmania than female migrants. Ordinal logistic regression was conducted using the GENLIN procedure within SPSS version 20.0 to determine the relationship between variables. The 95% confidence intervals were also calculated to test the significance between variables. Results were considered statistically significant at p ≤0.05 (Munro, 2005). All the data assumption tests including the test for multicollinearity were conducted before the ordinal logistic regression analysis. The results showed that no assumptions were violated.