This is strange. Trafalgar has produced 15 polls of Senate races since August 17th (you can find them by searching the linked polling database for “Trafalgar”). They love their tenths-of-a-percentage point “accuracy”. But not one of them has a single percentage with a tenths place of zero. That’s for any of the candidates, or for “other”, or for “undecided”.
This sample has sixty numbers. By sheer chance some of them should end in zero. If we assume that the tenth digit of each is randomly chosen (almost true, maybe not quite but the dependence isn’t a major factor), the chance that none of them are zero is (0.9)^(60), which is 0.0018, or 0.18%. That’s pretty unlikely.
Unless, of course, it’s not random. Zero is pretty boring, after all.
Fun fact #2: those 15 Senate races? Every one of the races has candidate totals adding up to 100.0%. You say, wait, why is that weird? Well, there should be rounding error *sometimes* that leads to 99.9% or 100.1%. Specifically, for three-candidate races this error should happen a quarter of the time. For more candidates, it should happen more often. So the chance that all fifteen add up to 100 — even if they were all three-candidate races (not true), it would be (0.75)^(15), which is 0.013, or 1.3%. Also unlikely. Not as unlikely, but still unlikely!
This interesting coincidence doesn’t seem to be replicated in their other data columns — for example it’s quite frequent that their age and/or ethnicity percentages don’t add to 100.0%, and it happens in some of the Governor’s races and even in party identification.
Also: neither of these problems seem to affect Trafalgar’s earlier polls — plenty of terminal zeroes in percentages (see Beasley-Budd from June), and plenty of failure to add to 100 (see the pre-primary MO polls among others). They turned on the fire hose on August 17th in terms of polling volume, and that’s when the strangeness starts.
I want to be very clear that none of this is proof of wrongdoing on Trafalgar’s part. I can see plausible explanations for fun fact #2 — maybe they adjusted the percentages to add to 100%. Sometimes people do this. (If so, they didn’t do so on the other columns, for whatever reason).
I don’t understand how the absence of zeroes could happen, though. That’s really suspicious, and it forces me to consider the possibility of data manipulation. Since extraordinary claims such as faking data require extraordinary proof, I am open to hearing alternative explanations.