Better vote sampling would have cast more doubt ~ above the potential because that Hillary Clinton to win the 2016 election

Donald Trump’s 2016 election win took numerous by surprised – many of the polling had said a win for Hillary Clinton. However were the polls wrong? In brand-new research Manfred dare Grotenhuis, Subu Subramanian, Rense Nieuwenhuis, Ben Pelzer and plunder Eisinga study the choice polls’ accuracy through randomly sampling from each state’s observed voter for Clinton or Trump. They discover that a reasonably small polling prejudice which experienced Republicans underrepresented in a number of key states tipped the polling – and also therefore the guess probability that she would win – in donate of Hillary Clinton.

You are watching: Chances of hillary clinton winning the election


Ahead of choice Day in 2016, the statistics and also polling analysis site FiveThirtyEight gave former Secretary the State Hillary Clinton a 70 percent probability to win the presidential election. This prediction was based on an median of numerous US polls. Follow to the American Association for Public Opinion study (AAPOR) this prediction turned the end to be incorrect due to the fact that of a) a real change in voters’ preferences just before the election, b) an overrepresentation of university graduates in part poll samples and c) late-revealing trumped voters.

What has been generally overlooked is the the polls leading approximately the election in reality did not do so badly at all. For instance, the state-level predictions of FiveThirtyEight were in agreement with the actual electoral outcomes because that no much less than 45 US says plus the ar of Columbia. If we sum all electoral votes in these states we obtain a essentially neck and also neck an outcome of 231 votes for Trump and 232 because that Clinton. In Florida, Michigan, Pennsylvania, Wisconsin, and North Carolina FiveThirtyEight had suspect wrongly. Personally from north Carolina, in these says the electoral spare part were incredibly narrow. For instance in Michigan 47.5 percent of all votes checked out Trump and also 47.3 percent to Clinton! Those small electoral margins more than likely made it difficult to guess the outcome in a reliable means given the sample size polls used. This is vital to note because the 2016 united state presidential choice was won in Florida, Michigan, Pennsylvania, and also Wisconsin v their decisive complete of 75 electoral votes.

In a perfect world, polls sample native the population of voters, who would state their political preference perfectly clearly and then poll accordingly. However, results from tiny random samples deserve to be fairly unreliable as result of extremely narrow electoral margins. To calculate the probability of winning the 2016 united state presidential election in that perfect world, we drew 1 million random samples (with a reasonable sample size of 1,500) from every of the four key state’s observed votes for Clinton, Trump, and also other candidates. Next, we counted the variety of random samples with the most votes because that Trump and the variety of samples through Clinton together the winner.

In Florida, about 677 the end of every 1,000 random samples had Trump as a winner, 314 samples favorited Clinton, and also 9 samples turned out to be inconclusive. Beside Florida, we calculated the probability the a Clinton win in Michigan in ~ 46 percent, and also both Pennsylvania and also Wisconsin in ~ 39 percent, again after drawing a million arbitrarily samples every state from the known population of voters. Recall the Clinton currently had 232 votes indigenous the other states + DC, and thus needed an additional 38 electoral votes to end up being the an initial female united state president. This implies she had actually to win Florida to add at the very least one of the various other three battleground says or success in Michigan, Pennsylvania, and also Wisconsin. Come illustrate: Hillary Clinton had a 2 percent (.314 x .463 x .387 x .385) guess probability of winning all 4 states. The probability, then, of to win the 2016 presidential choice is the sum of all eight to win combinations and also amounted to 29 percent. For Donald Trump, the eight paths to victory included up come 67 percent. To acquire a notion of how heavily the suspect probabilities count upon sample size, we calculated the probabilities for Clinton to success for sample sizes between 100 and 5,000 (see figure 1).

Figure 1The approximated probability the a Clinton victory utilizing random samples from the really 2016 us presidential election results


The approximated probability because that Clinton to win is no much greater than 35 percent in the graph above. V random samples the 1,500 voters per state that is about 30 percent. This number suggests that on average we will erroneously predict Clinton together the winner of the 2016 united state election 3 out of 10 times as soon as random samples the 1,500 space used. ~ above the communication of these calculations, the narrow 2016 electoral spare part in the 4 battleground says are not a huge threat come the validity the the polls’ predictions.


“#clinton as #trump” by Oli Goldsmith is license is granted under CC through SA 2.0

Next us investigate the result of little sampling bias on the polling results in the four crucial battleground states. Clinton was predicted to victory the renowned vote (i.e. The total variety of votes) through a distinction of about 3 percent if in fact this was close come 2 percent. Under the presumption that this 1 percent predisposition is state-independent, we included this number come the yes, really electoral outcomes. Come illustrate: in Pennsylvania Clinton received 47.9 percent of the votes and also Trump 48.6 percent. So we enhanced the population of Democrat voters to 47.9 + 1 = 48.9 percent. Subsequently the populace of Republication voter was diminished to 48.6 1 = 47.6 percent. Next, us randomly attracted 1 million random samples per state from this 1 percent biased populations and recalculated the overall probability to victory the elections. This rather tiny sampling bias made the predicted probability the 70 percent because that Trump come win readjust into a 30 percent win, a prediction more or much less in line with many polls’ suspect just before Election Day.

Next, us calculated the opportunities of a victory for Clinton for all sample sizes between 100 and also 5000, after taking right into account a 1 percent bias. As deserve to be seen in number 2, the not correct prediction rapidly increased as the sample size rose. So, with big samples practically in 9 out of 10 time Clinton is faelafilador.netly suspect to it is in the winner of the elections. This renders perfect sense: the 1 percent predisposition in democratic voters is detected fine by these large samples, in turn Clinton obtained the favorable odds of win the elections.

Figure 2 – The approximated probability the a Clinton victory utilizing random samples attracted from the 2016 us presidential election results + a 1 percent democratic bias.


The prominence of high quality poll samples when margins room narrow

In the last united state presidential choice polls handed Clinton a fair chance to victory the elections. The explanation because that this mishap probably is not little sample sizes or too narrow electoral margins together such. The point is that the polls very likely sampled native a populace that simply was no sufficiently representing Republicans. In many states that small representation bias was of no prestige as the electoral spare were large enough. However, in four crucial battleground states with a large reservoir the electoral votes, spare were an extremely narrow and permitted the tiny bias to dramatically tip the scale and made the polls suspect the overall odds in donate of Clinton.

When things get this chop again, high top quality samples are needed, which are far an ext representative the the united state voters 보다 the 2016 samples. However, also highly representative samples can not repair the predisposition of late transforms in politics preferences and late-revealing. Castle will constantly be a snapshot of a populace that for some component has not completely made the mind increase yet. We have to realize that political polls at ideal are a same prediction in ~ some allude in time and may constitute a wrong forecast of the actual election result, just because one candidate may have a larger share of late-revealers than the other. With this in mental both pollsters and also the media may have been a little too self-assured once they presented the outcomes of yet one more US 2016 choice poll.

Please review our comments policy prior to commenting.

Note: This write-up gives the views of the author, and not the place of USAPP – American Politics and also Policy, no one the London school of Economics.

Shortened URL because that this post:


About the authors 

Manfred te Grotenhuis – Radboud UniversityManfred dare Grotenhuis is an combine professor of quantitative data evaluation at Radboud University and an affiliate the the Interuniversity facility for Social scientific research Theory and Methodology (ICS). The does research and teaching in inferential statistics, age-period-cohort models, multilevel modeling, event history analysis, and also SPSS syntax. 

Subu Subramanian – Harvard UniversityS V Subramanian (“Subu”) is a Professor of populace Health and Geography in ~ Harvard University, and Director the a University-wide to plan on used Quantitative methods in society Sciences. That was also the establishing Director that Graduate studies for the interdisciplinary PhD regime in populace Health Sciences.

Rense Nieuwenhuis – Swedish Institute because that Social ResearchRense Nieuwenhuis is a is one assistant professor at the swedish Institute for Social research study (SOFI). He is a quantitative sociologist interested in how the interplay between social policies and demographic trends gives rise to economic inequalities. 

Ben Pelzer – Radboud UniversityBen Pelzer is an Assistant Professor the Quantitative Research methods at the department of Sociology / Social science Research techniques of Radboud University.

See more: The Future' Delorean For Sale Back To The Future Cars For Sale Online


Rob Eisinga – Radboud UniversityRob Eisinga is a professor of quantitative research techniques at Radboud University. His substantive interests worry the analysis of social and also political change, consisting of electoral and spiritual behavior, and also the excessive weight epidemic. His current methodological interest is in the evaluation of location data and also their null distribution in particular.