# Why mixed models and country fixed effects yield opposing significant findings?

I am looking at how differences in earnings change between individuals with different educational groups with every unit increase in macro unemployment rate. I use cross-sectional data between 2000 and 2015 in 25 countries. I compare two models, one that is constructed using mixed models and the other using country fixed effect. Both models show that differences in earnings increase between the high and low educated when macro unemployment rate goes up. However, they oppose when we compare the high and middle level of education. The mixed models show that earnings differences increase, while the country fixed effect method shows that earnings differences decrease when macro unemployment rate goes up.

I am a bit puzzled by these findings, since initially I thought that they should not be that different across the mixed model and the country fixed effects. I was wondering whether I was doing a technical error and I am trying to understand why these differences occur?

Here are the codes and findings:

Mixed Model: Model1: lmer(ln_wage ~ age + gender+ country_birth + macro_unemployment*education + year + (1| country_year)+ (education|country), dat=df)

Country fixed effect model: Model2= plm (ln_wage ~ age + gender+ country_birth + macro_unemployment*education + year, index=c("country"), model="within", data = df)