Dear list,
I wonder how to best calculate the effect size of an interaction term in a fixed effects model.

Dawson (2014) suggests to calculate f2 = (R2_2 - R2_1) / (1 - R2_2) where "R"_1 and R2_2 represent the variance explained by the models including and excluding the interaction term, respectively"

Three questions:
1) The formula seems strange to me, because it would create a negative f2 when the interaction term actually adds explanatory power. Is this just a mix-up in the paper and R2_1 and R2_2 should be switched?
2) I am not sure which R2 to use when running an xtreg, fe model. Is the "overall" R2 the correct one?
3) R2 values in fixed effects models tend to be very small -- so any effect size of a moderator would be very small as well. Is f2 still an appropriate measure then? (given that the effects would always look tiny when compared to the traditional benchmarks of small/medium/large effects)

Best regards,
Michael

Dawson, J. F. (2014). Moderation in management research: What, why, when, and how. Journal of Business and Psychology, 29(1), 1-19.