Respondent, country, and year fixed effects.

I have a nested dataset of a panel of respondents from several countries, varying over time (respondents can appear more than once in the data). My independent variable is at the country-year level, and my DV is a binary variable at the respondent level. I’m running a logistic regression with SEs clustered at the country level (because that’s where treatment is assigned). I’m debating what is most appropriate - country and year fixed effects, respondent fixed effects, OR country, respondent, and year fixed effects.

I guess that technically respondent fixed effects should include country and year differences (because respondents vary between and within countries + years). However, I get a different result when I use three way fixed effects (why?)