Different coefficients before and after adding year-fixed effect

Question

Initially, the coefficient of lnFDI was positive. However, after adding year FE, it became negative. Both were statistically significant. Could someone help me out? I really don't know how to explain or address this problem. Thank you

Answer 1

Should be the issue of the Simpson Paradox as you run the regression without fixed effect.

In the figure below, you can see that without fixed effect of gender, then the coefficient is positive. However, once you account for the fixed effect (gender in this example) then it reveal the true negative coefficients here

Back to your issue, your model without FE probably having Simpson Paradox issue here. My suggestion is to run two ways fixed effect, both for Industry and Year. It is a must, otherwise, you will be questioned for not adding two ways fixed effect here