Small Business Doctor

Public opinion polling has always suffered from some methodological problems, which is why economists have for decades preferred revealed preferences above stated preferences as more reliable data. Watch what they do, not what they say.

Aside from that issue, there is the perennial problem of ensuring that your sample accurately reflects your population. Random sampling is the usual solution to this problem. This problem, of course, is not unique to opinion polling, but opinion polling adds a wrinkle to this.

Not every person is equally willing to share their opinion with a stranger. Some people are eager to share their opinions with anyone who will listen. Others will only openly share their opinions with others they trust or with the caveat that the opinion not be made public. Still others need to have their opinions pried out of them with a lot of coaxing.

So, those who do respond to a polling firm’s request for their opinion are more likely to be within that group of people who are more willing to share their opinion with a stranger. That has always been true, and it has been dealt with by assuming that this difference in opinion sharing is not a relevant difference that would meaningfully bias the sample away from representing the population.

Whether this assumption was ever valid, it is almost certainly not valid now. Why? Because these days, to get enough respondents for their poll, polling firms need to send random people a text with a request to fill out their opinion survey. Or, worse, they try to call that random person.

What does that sound like? Spam and robocalls, which are about as popular as cancer these days.

Unsolicited communications should always and almost always do go straight to the trash. We have learned that unsolicited communications are almost always some form of a scam, and it is never worthwhile to sift through it all to find something worth paying attention to. These days there are finally easy ways to screen out these annoying and potentially dangerous unsolicited communications.

That’s why polling firms now need to pay people a lot more than they used to for stating their opinion. Yesterday, I was promised $20 by Mercury Opinion for sharing my opinion with them. By the way, I looked and Mercury Opinion appears to have a solid reputation as a polling firm, and I am not calling them out. In any case, $20 is a lot more than I would have guessed for what it would take to induce someone to share their opinion until it dawned on me that the rise of the attention economy and the ubiquity of harms that can befall your phone and your wallet from clicking on a shady link or taking that robocall mean that almost everyone heeds the sound advice to block and delete that stuff.

Almost everyone. Who clicks on a weird shortened link in a text message sent from some unknown number with a promise to pay you with a gift card for giving your opinion? Who gives some unknown Nigerian prince their bank account details?

I can think of three types of people who would, unlike myself, affirmatively respond to the spam text from Mercury Opinion. First and probably foremost are financially precarious folks who are indeed willing to risk falling prey to a scam for the promise of $20. Second are lonely people or loudmouths who are only all too willing to potentially fall prey to a scam with the expectation that they will be listened to. Third are people who simply do not know any better than to answer any question or click on any link that is directed to them.

If I’m right, then these are essentially the only people left who are now sharing their opinions with the polling firms. That is not a representative sample of the population. Whether it is economic polling to gauge consumer sentiment or political polling to gauge voter preferences, the person who would risk blowing up their phone or their bank account for the promise of a $20 gift card is not at all representative of the population at large.

That is why, as the Kansas City Fed found in a 2025 study, polls about consumer sentiment are essentially useless for forecasting consumer spending. At one time, before the Internet, there was some value but no longer. Here’s the money quote:

Recent data suggest consumer sentiment has been declining for the past several months, signaling a potential slowdown in spending. However, most measures of actual spending, such as core retail sales and PCE, have remained relatively stable. This discrepancy raises the question of how useful consumer attitudes are in predicting actual spending.

Polls are widely publicized and yet are largely useless. I can’t foresee them becoming any better going forward. It’s not like spam, robocalls, and online scams are going away anytime soon. The majority is becoming more silent as the spams and the scams grow.