for my data, "personal growth" construct correlates positively and significantly with "life satisfaction" (LS) (over .4) and it has higher correlations with "gratitude" (.6) and "meaning in life" (.6). In my regression model (with LS as outcome), "personal growth" has a negative and significant beta. If I keep personal growth in the model, beta for "gratitude" is significant, if I omit "personal growth", "gratitude" coefficient becomes non significant. How can I interpret this? Do I have a suppressor?
I appreciate your thoughts! Thanks
Thanks Dragos, personal growth was operationalised as the scores for Ryff's Personal Growth scale from Ryff's PWB scales. it includes questions about both desired and achieved.
I thought of the same explanation though. I think LS is not enough to measure wellbeing, maybe high LS preserves the status quo. Personal growth results, mostly, from overcoming negative events, so when you think of satisfaction, if you had a fair share of negative events lately, the answer for LS will be lower but maybe the score for personal growth will be higher.
Anyway, how should I explain the significance of gratitude when personal growth is included in the model? And how shall I explain the positive correlation? In statistical terms!
Many thanks!
What is the VIF for your factors in the data?
Suppose for a moment that I have a data set that consists of Weight, Height and BMI vs age (which goes from 1yr to 25yrs) . BMI is calculated from Weight and Height. If I look at the correlation for these factors vs Age, they should all be positively correlated, If I run a multiple linear regression, I will find that they are highly correlated with each other which gives a high VIF. That means the coefficients can show up as "significant" even thought they are not. Other coefficients can show up as positive, when its really negative, or negative when they are really positive, or insignificant if they are significant.
When you hear someone talk about a independent variables, part of that independence is a low level of correlation between each variable.
thank you Andrew! I understand that concept. I don't have a multicollinearity problem (highest VIF under 2.5). There are correlations between predictors but the multiple regression assumptions are not violated. I am more interested to know if you think that it is a suppression effect. Thanks
- Badges
- Science topic