The biggest issue is that you collapsed nuanced 5-point Likert responses (‘much better at home’ to ‘much better at office’) into binary variables before running logistic regression models. This means all middle-ground responses (e.g., ‘about the same’ or ‘slightly better’) were forced into the same category as an extreme. That wipes out a ton of meaningful variation and makes your logistic regression results pretty shaky. So are your ‘top predictors’ really trustworthy? It feels like they may reflect how the data was coded more than how workplaces actually affect performance or well-being.
