I'd seen this first in Adda and Cornaglia. But, like with any Peltzman effect, you have to check whether the worse outcome per instance outweighs the drop in instances. Drivers wearing seatbelts take more risks, but they are more likely to survive a crash. The ultimate effect is then ambiguous (though less ambiguous for pedestrians). It's an empirical question.
The latest AEJ: Policy has new evidence from Cotti, Nesson and Tefft. Their abstract:
We analyze the Nielsen Household Consumer Panel to estimate the effects of tobacco policies on tobacco-related purchases using within-household variation. We also match purchases to cigarette contents from NHANES. Higher cigarette taxes reduce cigarette purchases and increase smoking cessation product purchases, while estimates of smoking ban effects are less precisely estimated. Smokeless tobacco (SLT) taxes lead to reductions in SLT use but also lead to substitution among SLT products. We find evidence that cigarette taxes induce purchases of cigarettes with higher tar, nicotine, and carbon monoxide contents, but this compensatory behavior is overwhelmed by the reduction in cigarettes purchased.So more harms per cigarette (small effect), but dwarfed by the reduction in cigarettes smoked.
Unfortunately, their "smoking cessation product purchases" category includes both standard nicotine gums and the like, and electronic cigarettes. It would have been very interesting to see how e-cigarette availability mediated the effects of tax hikes.