Hello Everyone,

I would greatly appreciate some help with this question. I am trying to help a good friend with her dissertation and am now confused and unable to help her.

Here is the background of the study:

Her dissertation is examining whether race/ethnicity status predicts perceived stigma scores for breastfeeding in public and whether several variable are moderators between the IV and DV. She has 3 separate hypotheses.

1. Does Participant Education Level moderate the relationship between race/ethnicity status and perceived stigma scores?
IV: Race/Ethnicity Status (categorical, dummy coded for white and non-white)
DV: Perceived Stigma Scores (ordinal- total score from Likert-type measure)
Moderator: Education Level (ordinal- participants assigned numbers for completing different years of education 1-junior high, 2-some high school, 3-high school/equivalent ect).

2. Does social support moderate the relationship between race/ethnicity status and perceived stigma scores?
IV: Race/Ethnicity Status (categorical, dummy coded for white and non-white)
DV: Perceived Stigma Scores (ordinal- total score from Likert-type measure)
Moderator: Social Support- (ordinal- total score from Likert-type measure)

3. Does prior breastfeeding knowledge moderate the relationship between race/ethnicity status and perceived stigma scores?
IV: Race/Ethnicity Status (categorical, dummy coded for white and non-white)
DV: Perceived Stigma Scores (ordinal- total score from Likert-type measure)
Moderator: Prior Breastfeeding Knowledge- (ordinal- total score from Likert-type measure)

Problem/Question:

She had 7x more white participants than non-white participants and very different sample sizes of white and non-white. This resulted in a violation of her assumption of homogeneity of variances. She then looked at oversampling the minorities a suggestion from Aguinis. This still violated the assumption of homogeneity of variances. So she then decided to randomly remove some white participant in SPSS to make her sample equal number of white and non-white participants. She then met the assumption of homogeneity of variances...YAY!!!

However, now her advisor has asked her is she should have done a paired/matched sample of the white participants to the non-white participants based on other demographic variables (participant's education level, partner's education level, age, marital status) and why or why not. We do not know the answer and have yet to find it in stats books or online. Can someone please either provide the answer or give me a reference to find the answer? Thank you for you help and sorry this is so long