Perhaps you have wondered how the inclusion of the Green and Libertarian Party candidates, Gary Johnson and Jill Stein, is affecting the race between Hillary Clinton and Donald Trump. It is something I have wondered about for some time so I finally decided to try analyze that situation. This is this methodology I devised.
With two 3rd party candidates siphoning off votes from the both of the two major party candidates it is difficult to determine whether, for example, Johnson helping or hurting Hillary? Or whether Johnson is helping or hurting Trump? However, we can take some examples of what I will call “paired polls” of the national race and those in some of the states, make some common sense assumptions, and come to some reasonable conclusions about who is hurting or helping who.
When I talk about “paired polls” I am referring to situations where polling organizations ask their sample populations their preferences in both the two person race and the four person race.
Example 1:
The following are the results of a set of paired polls for the State of Pennsylvania conducted by Suffolk University between 7/25 and 7/27:
Clinton 44, Trump 44, Tie
Clinton 43, Trump 39, Johnson 5, Stein 1 Clinton +4
Notice that when the 3rd party candidates are added, Clinton’s percentage drops by 1% (44% – 43%) and Trump’s percentage drops by 5% (44% – 39%). Already we can see that Hillary does better against Trump in the 4 person race, but how do we figure out which of the third party candidates is helping her the most? Now let’s make a common sense assumption – that Stein is siphoning off votes only from Clinton in a 4 person race. (Rational; It seems highly unlikely people who vote for Stein in a 4 person would vote for Trump in a 2 person race. I could see Stein voters sitting out a two person race or writing in a candidate, but I can’t see them voting for Trump. This might not be a perfect assumption, but it good enough for our purposes.) Note the total percentage of voters in each of the two types of races both add up to 88%. So it doesn’t look like any voters are moving to the undecided or voting other columns when moving from the 4 person race to the 2 person race so this factor can be ignored.
Proceeding with our assumption, we add Stine’s voters to Hillary’s to reduce the 4 person race to a imaginary three person race:
Clinton 43, Trump 39, Johnson 5, Stein 1 Clinton +4 (4 person race)
To: Clinton 44, Trump 39, Johnson 5 Clinton +5 (3 person race)
Now let’s compare the three person race (above) the two person race (Clinton and Trump):
Clinton 44, Trump 44, Tie
Clinton 44, Trump 39, Johnson 5 Clinton +5
When Johnson is included, Clinton’s total decreases none at all (44% to 44%) while Trump’s total decreases by 5% (44% to 39%)
Conclusions: In this set of paired polls, Johnson’s presence in the race hurts only Trump when we assume that Stein hurts only Clinton with her candidacy. We can also see that Johnson hurts Trump far more than Stein hurts Clinton.
Now lets go though some additional paired poll examples:
Example 2:
One of the last paired polls for the national race was on ducted by CNN/ORC between 7/18 -7/26.
Trump 48, Clinton 45 Trump +3
Trump 44, Clinton 39, Johnson 9, Stein 3 Trump +5
Notice that when the 3rd party candidates are added, Trump drops by 4% and Clinton’s percentage drops by 6%. However, if you add up the percentage votes in the 4 person race it comes up to 95% while the total votes in the two person race is only 93%. Apparently 2% of the voters in the four person race have dropped back to undecided or decided to write in a candidate. This is difficult to handle, but just to be able to proceed I am going to assume that 1% of that 2% total were Johnson voters and 1% Stein voters. So to make the two polls compatible in total number voters I have subtracted 1% from both Johnson and Stein in the 4 person race:
Trump 48, Clinton 45 Trump +3
Trump 44, Clinton 39, Johnson 8, Stein 2 Trump +5
Now let’s make our common sense assumption – that Stein is siphoning off votes only from Clinton. Adding Stine’s votes to Hillary’s for a three person race:
Trump 44, Clinton 41, Johnson 8, Trump +3
Now let’s compare the two person race (Clinton and Trump) to the three person race (Clinton, Trump and Johnson).
Trump 48, Clinton 45 Trump +3
Trump 44, Clinton 41, Johnson 8, Trump +3
When Johnson is included, Clinton’s total decreases by 4% (45% to 41%) while Trump’s total decreases by 4% (48% to 44%)
Conclusions: If our assumptions are correct, in this case Johnson is pulling voters equally from both Clinton and Trump, but Stein hurts only Clinton with her candidacy.
Example 3:
This is a similar set of paired polls for a national election, by Reuters/Ipso between 7/16 and 7/20
Clinton 44, Trump 44, Tie
Clinton 43, Trump 39, Johnson 5, Stein 1 Clinton +4
Notice that when the 3rd party candidates are added, Clinton’s percentage drops by 1% (44% – 43%) and Trump drops by 5% (44% – 39%). The total votes in both the 2 and 4 person races add up to 88%, so we don’t have to worry about a situation where voters in one race are not participating in the other race. Now let’s make our common sense assumption – that Stein is siphoning off votes only from Clinton. Adding Stine’s vote to Hillary’s for a three person race:
Clinton 44, Trump 39, Johnson 5 Clinton +5
Now let’s compare the two person race (Clinton and Trump) to the three person race (Clinton, Trump and Johnson).
Clinton 44, Trump 44, Tie
Clinton 44, Trump 39, Johnson 5 Clinton +5
When Johnson is included, Clinton’s total decreases none at all (44% to 44%) while Trump’s total decreases by 5% (44% to 59%)
Conclusions: Assuming our assumption is correct, Johnson hurts Trump far more than Clinton with his candidacy – in fact all of Johnson’s votes are coming out of Trump’s hide. Johnson helps Clinton beat Trump while Stein is only negative a factor for Clinton.
Summery: I think now you can see the basic methodology now which you can apply easily to any paired race in the future. In fact, if you get really familiar with the basic technique and you are good with numbers you probably will get to the point where you can do the analysis of a paired poll in your head.
Notice that the results will vary from race to race so we can draw no over arching assumptions from just the three samples shown here. If we look only at the three examples we might come away with the conclusion that the inclusion of Johnson into the race is helping Clinton more than Trump, but that overall conclusion is premature. One would have to look at many examples to draw a valid conclusion.
One thing to remember however, whenever the total percentage of voters in the four person poll is greater than percentage of voters in the two person race (see Example 2), you have to do some guessing as to whether Johnson’s voters or Stein’s voters or some of both are sitting out the 2 person race. This hurts the perceived accuracy of the analysis; in my opinion the greater the differences of those percentages between polls, the more you are guessing and the more error can creep in to your work product.
Cajun 7/30/16