Rainfall variability in Sub-Saharan Africa creates significant production risks for subsistence farmers who rely on rainfed agriculture. Weather shocks can alter farmers' risk preferences, potentially influencing their decision to adopt adaptation strategies. We employ a moments-based approach (Antle 1983; 1987) to estimate risk aversion among smallholder farmers in Ethiopia, Malawi, and Nigeria in response to weather shocks. Using this framework, we estimate Arrow-Pratt and downside risk coefficients across more than 200 model specifications, incorporating various rainfall and rainfall shock metrics derived from six remote sensing weather products. Our findings reveal that estimates of farmers' risk preferences are highly sensitive to the choice of weather product, while risk preferences are relatively insensitive to different weather shocks. This underscores how data source and model specification choices critically shape risk preference estimates, highlighting key limitations in current empirical methods.