I experienced two major roadblocks in this analysis. First, I originally planned to analyze the context of each science reference. Were these references positive or negative? Unfortunately, if several science or religious references appeared within thirty characters of each other, the surrounding words were appended to the list more than once:
String: I think we should study more science, math and evolution.
My list: ['I think we should study more science, math and evolution', 'I think we should study more science, math and evolution', 'I think we should study more science, math and evolution']
In this instance, the word 'think' is counted as "surrounding text" three times, which produces false results if I simply compared the above list to a list of positive or doubtful words (or positive/negative, etc.). Instead, I ended up printing the context to each statement and analyzing the average and standard deviation for every debate.
Second, I was unable to conclusively support or disprove my original claim due to extremely high standard deviations. As such, further testing is warranted, including examining more debates including general election debates and older / newer transcripts as well.
Both of these roadblocks ended up beyond the scope of this particular project.