If you think Trump is losing, you’re in for a nasty surprise

Polls pointing to a Kamala Harris victory appear to be skewed by including too many Democrat voters

Looking at some of the polls in the race for the White House, you would be forgiven for thinking the race was over. Most show a lead for Kamala Harris over Donald Trump. Some polls last week showed her ahead by five, six, even seven points.
If this sounds familiar, it is because this happened in 2016 and 2020. At this point in the 2016 cycle, Hillary Clinton was ahead by 4 points; she went on to win the popular vote by half that and lose the Electoral College to Donald Trump. An ABC News poll on the eve of the 2016 presidential election had Hillary Clinton ahead by 12 points.
And at this point in 2020, Joe Biden was ahead by 8 points; he went on to win by 4.5 points. The polls were off then, and the signs are the same thing is happening again. 
The analyst Ryan Girdusky has spotted one of the problems. Voters over the age of 65 backed Trump in the last two elections, but polls are now indicating Harris is going to beat Trump decisively amongst that age group. 
Rather than that being a genuine shift, Girdusky has identified that older white liberals are more likely to answer polls than older white or Hispanic conservatives. That response bias – answering polls at a far greater frequency than other groups – skewed the polls in the last two elections and now the same is happening again.
There are other problems, too. The surveys showing Harris the farthest ahead are polls that are conducted entirely online.
That is normal practice in the UK. And although polls over-stated Labour’s lead at this July’s election, the best-calibrated online polls were able to get close to the overall result. They were accurate in 2019 as well.
The same cannot be said for the US, where whether a voter completes an online survey differs hugely depending on how that voter looks and thinks.
By digging into the data underneath the Harris leads, we see those online polls over-sample three key groups: younger Americans, voters who are the most engaged in politics, and people who work remotely from home during the day. 
All those groups are more likely to be Democrat. 
Add in the fact that older Republicans use the internet less and are more likely to distrust entering their personal data online, and it is clear that there is a problem.
A counterpoint made by defenders of the polls is that they were accurate in the midterm elections of 2018 and 2022. They predicted a good Democrat performance, and they got it right. 
But Trump was not on the ballot at those elections. Because Trump voters are less likely to be politically engaged (seven in 10 of them are non-graduates, a good proxy for political engagement, compared to just half of Harris’s support), his inclusion means that there are more voters that online polls struggle to reach.
The good news is that some polls are more accurate than others. The most effective way to survey the US electorate is through mixed method polls, that combine online surveys with phone calls to landlines, people’s mobiles, and via SMS.
I learnt this lesson the hard way in 2020. Thinking that we could replicate the success of online methods in the UK across the pond, we polled Biden versus Trump. We registered a 10-point landslide win for Joe Biden.
Since then, we have changed our methods and now utilise a mixed method approach. Along with calls and texts, we also use in-app surveying: if you are playing a game on your phone, you might get a notification asking you to complete a survey to unlock more in-game points. This picks up people who might never otherwise answer the phone to a pollster.
This approach allowed us to get the most accurate result in the country in the JD Vance Senate race two years ago. The real test is to come this year. Our national polling currently shows a 2-point lead for Donald Trump over Kamala Harris. 
Other similar methods by other polling outfits show a strong performance for Trump or have him pushing Harris to a tie.
Polling still has its uses. But if you look at just the online ones – which pro-Kamala Democrats keep pointing at – all signs are that you are headed for a nasty surprise come November.
James Johnson is the co-founder of JLPartners and was a pollster and strategy adviser for Theresa May