Quite a few reports about the latest political maneuvering in Washington DC include Trump’s latest polling numbers. This is usually followed by some amazed discussion of how his support amongst Republicans is so consistently high, the latest being 90%. What will it take? They ask, for Republicans to turn away from Trump? The answer to that remains unknown, but it is entirely possible that moderate Republicans have already turned away from Trump and the polls are not picking it up.  This could be happening even if the polling companies are following all of the proper practices of random polling.

Here is how random sample polling works.

A polling firm wants to estimate Trump’s support amongst the general public, Republicans, Democrats and Independents in a state with an electorate that is 45% registered Republican, 45% registered Democrat and 10% unregistered or registered as Independent. They are looking for a high degree of accuracy so they want 10,000 respondents. They want a representative poll so they want 4,500 of those respondents to be Republicans, 4,500 to be Democrats and 1,000 to be Independent/unregistered.

The firm randomly generates a list of 20,000 phone numbers because they expect that up to 50% of the people they contact will refuse to participate. This is called the refusal rate. The polling firm does not know the political allegiance of the person answering the phone so if they agree to respond they ask for it. The firm trusts them to tell the truth but has no way to verify it.

The firm may obtain 1,000 responses from Independents long before they get to 4,500 responses from either party. At that point they stop polling Independents once the respondent identifies as such. Once they reach 4,500 for either party they will also stop polling self-identified members of that party and will continue to call until they get 4,500 from the other party. This is all done to try to get a representative sample of voters. They don’t know who will actually vote so they may also ask about voting intentions and they trust the respondent to tell the truth.

One explanation of Trump’s durable high polling amongst Republicans is that it is accurate and he actually has the support of up to 90% of registered Republicans. Another possible explanation is that fewer and fewer registered Republicans are self-identifying as such to pollsters and either not contributing to the poll or are counted as Independents. The polling firm in the above example could exhaust all of the remaining numbers on its list trying to get 4,500 Republican responses and may even have to generate more numbers.

Some organizations conduct tracking polls in which they call the same people on a regular basis. These polls are not random and are only representative if their respondent group is representative of the general public. Both the Democrats and the Republicans use their own party lists to poll their own members. They know whom they are calling.

The bottom line? When we see a poll saying anything, we really should be looking at how they collected their data in order to draw that conclusion. The key questions are: Were proper random polling practices followed? Do respondents self-identify as party members or are their affiliations known before they are called? If self-identified, what is the refusal rate for this poll and how does it compare to previous and historical refusal rates? Which party quota was reached first, which second? Most polling companies don’t release such information. It isn’t because they are trying to deceive us; they just think we don’t care.