Dear all,
I performed a linear multilevel regression analysis [including 21 countries], proceeding stepwise from null-model (M0), including the main explanatory variable discrimination (M1), individual-level controls (M2) and country-level controls (M3). Then, I added a random slope for discrimination (M4).
The significance level of discrimination lowers in M4 (p<.05) compared to M3 (p<.001), but the Likelihood Ratio Test suggests to choose the less parsimonious model M4.
The interpretation is: the impact of discrimination on my dependent variable seems to vary across the 21 countries I included.
Now my question: How can I compare the impact of discrimination across the countries?
I found one study [10.1371/journal.pone.0074252] that used M4, dichotomized the outcome variable (good vs. bad health) and created the following graph:
Would you say this is an appropriate strategy to evaluate the impact of an independent variable across different groups?
If yes, does anyone know how to create such a graph, controlling for the background variables of the model or is it just possible by performing a new, logistic multilevel regression because the outcome must be a dummy? Are there any suggestions for another solution?
Best regards,
Rebecca
I performed a linear multilevel regression analysis [including 21 countries], proceeding stepwise from null-model (M0), including the main explanatory variable discrimination (M1), individual-level controls (M2) and country-level controls (M3). Then, I added a random slope for discrimination (M4).
The significance level of discrimination lowers in M4 (p<.05) compared to M3 (p<.001), but the Likelihood Ratio Test suggests to choose the less parsimonious model M4.
The interpretation is: the impact of discrimination on my dependent variable seems to vary across the 21 countries I included.
Now my question: How can I compare the impact of discrimination across the countries?
I found one study [10.1371/journal.pone.0074252] that used M4, dichotomized the outcome variable (good vs. bad health) and created the following graph:
Would you say this is an appropriate strategy to evaluate the impact of an independent variable across different groups?
If yes, does anyone know how to create such a graph, controlling for the background variables of the model or is it just possible by performing a new, logistic multilevel regression because the outcome must be a dummy? Are there any suggestions for another solution?
Best regards,
Rebecca
