Explaining human sampling rates across different decision domains