Gravar-mail: Potential confounds in estimating trial-to-trial correlations between neuronal response and behavior using choice probabilities