At the IEA blog, Kristian Niemietz points out that expenditure surveys fail to detect most of the tobacco sales visible in national product accounts

For most goods, the two show broadly the same pattern: with small
errors, what people profess to buy grosses up to what is really being
sold in the country. But tobacco is a big exception. Less then half of
the recorded cigarette purchases shows up in the Living Cost and Food
Survey. In the US equivalent, the ratio is not even 40%.

Why is tobacco dark matterSocial Desirability Bias!

The mismatch between what smokers say in surveys and what they do in
practice is a classic example of the difference between “stated
preferences” and “revealed preferences“.
Social engineers love stated preferences. Opponents of big
supermarkets, too, always have a survey at hand, indicating that the
vast majority of residents in their areas would never set foot in a
discounter. But once it is there, it flourishes.

There is nothing schizophrenic about this behaviour. When asked
whether you would shop in a big supermarket in your area, of course you
respond something like “No! Small, local shops are much more charming and personal”
– because that is the socially acceptable thing to say. When you smoke,
saying that you want to quit makes you at least a repentant sinner.

Now ask yourself: Is voting more like a national product account – or a consumer expenditure survey?