Robustness analysis leads to different results. What to do now?

#1
Hello everyone,

My thesis is about the price determinants of Bitcoin. Using an ARDL I identified several explanatory variables. (Variables have been transformed to logarithmic return). One of the key explanatory variables is a proxy for public sentiment. So far, so good.

In my robustness section I changed the sentiment proxy for another one. In the main results I uses Thomson Reuters Marketpsych Sentiment and in the robustness section I made use of another data supplier. Some other explanatory variables turn insignificant, as indicated by the robustness analysis. How can I explain that in my thesis? Do you have any tips?

I already mentioned in the beginning of the robustness section that the additional sentiment index contains less observations (1700 vs 2500), is probably based on different sources and uses a different scale (-5/+5 vs -1/+1).

Any help is appreciated!