Hi @jessegrabowski,
Thank you for this very thorough response. I modeled temperature as a mixture because my observed data is distributed with three distinct peaks. A unimodal Normal doesn’t seem to fit the data as well as the mixture, as seen in the graphic below.
However, doing so certainly brings the result more in line with the regression result (see traceplots below), but does not fit the data as well.
As you’ve alluded to, this temperature data is aggregated across various years and regions, which may be the reason for the multimodality. However, I’m interested in the global effect of temperature on GDP across time, which is why I am using this observed data.
I’m wondering what negative downstream effects using a model fit that looks as “off” as the first figure will have. Even if the sampling result now agrees with the regression, isn’t this approach missing out on fully capturing the distribution of the data?