**Appendix C. Normalization of Weights**

To compare the relevance of the information vectors we need to transform the weights into comparable units. Therefore, we use a simple form of transformation by multiplying the estimated coefficients by the standard deviation of the data. The transformed coefficient would approximate how much a standard deviation in the data affects the similarity function. This does not change the interpretation of statistical significance, but does provide a way to assess the economic significance of the information vectors to subjects.

In Table A2, we find that the moving average captures more relative weight in constant sum games while recency is weighed more heavily in non-constant sum games. While both weights appear to be economically significant to define similarity of cases across all games, a standard deviation change in the moving average of opposing players garners much more weight on choices on average. Consistent with our previous interpretation, the non-constant sum games tend to discount cases in the relatively far past heavily compared to constant sum games.


**Table A2.** CBL Normalized weights.
