« Previous | Contents | Next »
Listen
ANNEX F : MULTIVARIATE ANALYSIS TECHNIQUES
F.1 Regression analysis is a technique used to identify factors that contribute to an outcome and to give an indication of the relative strength of these factors. The most common form of regression is linear regression, which is analogous to drawing a 'line of best fit' through the data. The regression process considers each one of the contributory factors under consideration and systematically tests combinations of those factors to find the collection of factors that most accurately defines the outcome. In the context of the 2006 survey, regression analysis was used to identify which variables have the strongest relationship with attitudes to mental ill-health.
F.2 Segmentation analysis is a way of simplifying survey questions into a smaller number of themes or 'factors' by grouping together items that are answered in similar ways. The process involves factor analysis to identify the common themes, followed by cluster analysis to segment the sample into groups based on these themes. The demographic composition of each of the typologies can then be analysed. Segmentation analysis was used to provide a fuller picture of responses to the attitudinal statements presented in the survey, including the extent to which responses vary among different groups of respondents.
F.3 Correlation analysis compares two variables and assesses to what extent (i.e. how strongly) they are related to each other. The best way to visualise correlation is in terms of a scatterplot. The x-axis represents one variable and the y-axis represents another variable. Each respondent has a score on variable 1 and a score on variable 2, so each respondent can be represented as a point on this scatterplot. The pattern of points formed by plotting each respondent will dictate the strength and direction of correlation. A strongly correlated pair of variables will form a pattern resembling a straight line. The orientation of the line will dictate the direction of the correlation. If the line slopes upwards (i.e. from bottom-left to top-right), then the correlation is positive. If the slope is downwards, then the correlation is negative. However, as with regression analysis, correlation analysis can identify only an association between variables; it cannot tell us if the association is causal, or the direction of causality.
F.4 The aggregation of responses across variables provides another means of generalising about the ways in which particular types of questions are answered by different groups of respondents. Typically, it involved allocating a score to respondents which reflects their pattern of responding across a number of questions, and then considering to what extent higher and lower scores are correlated with other attitudes and behaviours. In the present study, this approach was used in a number of different ways. For example, respondents were given a 'social engagement' score based on their answers to the questions concerning their informal support networks and civic participation. Similarly, they were given a 'social distance' score which reflected their responses to the battery of questions exploring their willingness to interact with someone displaying symptoms of mental ill-health.
« Previous | Contents | Next »