OAR Comment: Correlation? Association?
Sunday, November 28th, 2010I am struggling with the OAR comments and have no chair/mentor to ask for clarification or help. Here is one of the comments:
Research Questions: Since the PSR dimensions that will be the dependent variables are described as “interrelated”, the Learner must include a discussion of how those interrelationships (i.e., correlations) will impact the assumptions of the selected statistical analysis procedures (and how the interrelationships will be controlled/handled if they represent a threat to those assumptions) as well as how those interrelationships might impact the interpretation of the findings.
The variables in the proposal are sustainability status (nominal/independent) and levels of five purchasing social responsibility (PSR) dimensions (ordinal/dependent). The five PSR dimensions are related because they are all part of socially responsible corporate procurement policies. As these are not continuous, from my readings it appears that I cannot determine correlation (as above statement implies) but rather only association. I plan to use MANOVA analysis to identify associations among variables, because the variables are related. I am not sure I understand exactly what the OAR is asking in reference to how the interrelationships will be controlled/handled if they represent a threat to those assumptions. Here is what I have written so far:
There are five assumptions that must be met for MANOVA: the sample must be random, observations must be independent, each variable must be normally distributed, and within-group homogeneity of variance should be the same for both groups (Bray & Maxwell, 1985). Violating the assumptions does not invalidate MANOVA; it reduces robustness (Bray & Maxwell, 1985). Before running MANOVA, data will be tested for normality and outliers, which will be removed. Homogeneity of variances will be computed by computing the error variance (SS error) by adding the sum of squares within each status group (French, Macedo, Poulsen, Waterson, & Yu, 2008). Wilks’ lambda will be used to test mean differences in the two groups based on the combinations of PSR dimensions. SPSS converts Wilk’s lambda to an F distribution; if lambda is significant, “it implies that there is at least one significant contrast between groups and this difference can be detected when the entire set of outcome variables is examined” (Warner, 2008, p. 715).
My chair resigned and I don’t have that resource to consult. I am struggling with understanding several of the OAR’s complex comments. After all, this is a concept paper only – not the dissertation proposal.