May 22, 2020
In early April, I statistically demonstrated that the coronavirus had no effect on Donald Trump’s Job Approval ratings (Has the Coronavirus Hurt Donald Trump’s Reelection Chances?). That post showed that pre-virus period approval ratings were not statistically different than the virus period.
Now I read and hear that Trump’s polling numbers against Biden are being hurt by his handling of the Covid crisis. If you google that phrase you will find dozens of articles making this observation.
“Perhaps more worrying for Trump are his dwindling numbers in a match-up against Joe Biden.” The Guardian
Admittedly, polls on how well Trump is handling the Covid-19 crisis have been largely negative. The internet site 538 has polling that shows that 51% disapprove of his handling of the coronavirus crisis. Consequently, it has to be hurting his polling numbers.
This assumption requires that the two concepts are identical. But a person chooses a candidate for many different reasons and is not always dependent on one issue alone. In other words, you could agree he has mishandled the covid crisis and still vote for him.
To test this hypothesis, I collected 80 different independent surveys that included the Biden vs. Trump trial ballot question, from September 2, 2019 to May 19, 2020. This time line allowed me to equally divide the polls into two groups: pre-virus period and the virus period.
Among all 80 polls, Biden averaged 49% and Trump 43%, a six point lead for Biden. When I divided the polls into two groups, Biden’s and Trump’s pre-virus average was 50.2% and 43.8% respectively, or a 6.4% difference.
Among the virus period polls, Biden’s poll average was 48.4% and Trump was 42.7%, or a 5.7% lead for Biden. That’s a lead difference of 0.7% between the two virus periods.
It should be obvious at this point that the two-time frame polls were on average almost identical. That, however, doesn’t necessarily mean that the two groups, pre-virus and virus periods, aren’t different. Our interest here is Donald Trump’s polling and not Biden’s.
To prove that Trump’s two group means (virus and pre-virus) are different or the same, we need to test the observed variance in this variable and partitioned into components attributable to different sources of variation, which in this case is the handling of the Covid-19 crisis.
To do this, I’ll use a statistical technique called analysis of the variance (ANOVA), to determine if we have two different data means or basically both are the same. If the virus crisis has hurt Donald Trump, we would expect the two groups (pre-virus and virus periods) to differ. If the groups are the same, the variation would be nearly identical.
Our null hypothesis is that there is no statistical difference between the two groups. The table below shows the output from the ANOVA comparison of the two groups. The only value you need to note, is the significance level (Sig.) which is .606 which is not significant.
|Sum of Squares||df||Mean Square||F||Sig.|
Since it is not significant (at the <.05 level) we can conclude that the null hypothesis is correct: there is no difference between the two groups and consequently, his handling of the crisis has not effected his polling against Biden.
A graphic representation of the two time periods and Donald Trump’s percent of the trial ballot question is shown below.
This graph tracks Trump’s virus and pre-virus percent in polls matched against Biden. His pre-virus average percent is 43.8% and during the virus period is 42.7%, a difference of 1.1%. Since the average margin of error for all surveys is +/-3.0%, this difference is just random noise.
So why do so many analysts believe the Covid virus is hurting his polling numbers? First, as I pointed out, many are relying on the single polling question of the voter’s rating of how well he is handling the crisis, which have been generally negative and automatically making the assumption it is hurting his Biden vs. Trump poll numbers.
In addition, there is tendency for people to focus on one or more survey results that show his numbers decl