# Lesson 8: Health Behaviors, Mental Health, and Social Influence

*Companion-podcast transcript • Sarah & Kiffer*  
*Office Hours episode to listen to after working through the lesson*

---

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. This episode goes with Lesson eight, Health Behaviors, Mental Health, and Social Influence.

**Sarah:** It's meant for after you've worked through the lesson. We'll bring some perspective on how these ideas get used in public health practice, some critique of the evidence behind them, a look at the questions students tend to find thorny in this material, and a few extra worked examples, including some that are harder than the ones on the lesson page.

**Kiffer:** There are a few places where we'll ask you to work something out before we give the answer. When that happens, you'll hear a few seconds of quiet. Pause the audio if you'd like more time to think it through.

**Sarah:** We have five questions. First, how Doll and Hill's numbers, which look so close together, could be strong evidence. Second, why quit programs that work still did less than taxes and laws to bring smoking down. Third, whether behaviour really spreads through friendships. Fourth, why strong evidence hasn't settled the argument over supervised consumption. And fifth, a question Kiffer and I see differently, whether loneliness is really comparable to smoking.

**Kiffer:** Question one. In Doll and Hill's 1950 study, among the men, 99.7 percent of the lung cancer patients smoked, and so did 95.8 percent of the hospital controls. Sarah, I'd like you to tell everyone what you make of those two numbers.

**Sarah:** Honestly, my first reaction is that they look almost the same. If I divide 99.7 by 95.8, I get about 1.04. So the cases were only about four percent more likely to be smokers. That sounds like a weak association to me.

**Kiffer:** That's the reaction a lot of people have, so let's test it. Smoking was so common among men at the time that even most of the controls smoked, so most of the information is hiding in the small group who didn't. Here are the actual counts. There were 649 men with lung cancer, and 647 of them smoked, which leaves two non-smokers. Among the 649 male controls, 622 smoked, which leaves twenty-seven non-smokers. So here's the question for everyone. Look only at the non-smokers. How many times more common was not smoking among the controls than among the cases? Take a few seconds.

*(Pause)*

**Sarah:** Twenty-seven non-smokers among the controls, and two among the cases. Twenty-seven divided by two is thirteen and a half. So not smoking was about thirteen and a half times more common among the men who didn't have lung cancer.

**Kiffer:** And that's the same information I divided away when I compared 99.7 with 95.8. Where did my first answer go wrong, do you think?

**Sarah:** When almost everyone is exposed, the percentages of smokers get squeezed up close to one hundred. Any two groups would look similar on that scale. The comparison that shows the size of the effect is the odds. Among the cases, the odds of smoking were 647 to two, about 324 to one. Among the controls, they were 622 to twenty-seven, about twenty-three to one.

**Kiffer:** Good. And what do you get when you divide the odds for the cases by the odds for the controls?

**Sarah:** About 324 divided by twenty-three, which is roughly fourteen. So the odds of smoking were about fourteen times higher among the lung cancer patients. That's a strong association, and it hid behind two percentages that looked nearly identical.

**Kiffer:** That's the general lesson. When an exposure is very common, ask about the people who escaped it. Now for some perspective on how this evidence was received. The best-known critic was the statistician Ronald Fisher. He proposed what he called a constitutional hypothesis, that some inherited trait might make a person both more likely to smoke and more likely to develop lung cancer. Fisher was also a consultant to the Tobacco Manufacturers' Standing Committee, a British tobacco industry body, which is worth knowing when you weigh his objection.

**Sarah:** But the hidden-factor idea itself is a fair question. A shared cause can produce an association without any causal link.

**Kiffer:** It is fair, and the reply to it was clever. In 1959, Jerome Cornfield and colleagues worked out how strong a hidden factor would have to be. Their reasoning was that if smokers have about nine times the lung cancer risk of non-smokers, any hidden factor that explained all of it would have to be at least nine times as common among smokers. Nobody could point to a trait that lopsided. So a very strong association is hard to explain away by simple confounding, which is part of why strength comes first among Hill's viewpoints.

**Sarah:** Question two. Theory-based quit programs work, with something like a five percentage point gain in quit rates at twelve months. Yet most of the drop in smoking came from taxes, smoke-free laws and advertising limits. If the programs work, why did they do so little at the population level?

**Kiffer:** Let's work it out with a made-up province, and I'll make it a two-step problem. The province has one million adults, and twenty percent of them smoke, so two hundred thousand smokers. It runs a quit line that adds five percentage points to the twelve-month quit rate, compared with quitting without help. In a typical year, five percent of the province's smokers call it. How many extra quitters does the quit line produce in a year? Take a few seconds.

*(Pause)*

**Sarah:** Okay. Five percentage points of two hundred thousand smokers is ten thousand extra quitters a year. That seems like quite a lot, actually.

**Kiffer:** Before you go further, I'd like you to check who actually receives that five point gain.

**Sarah:** Only the people who call. I applied it to every smoker in the province. Five percent of two hundred thousand is ten thousand callers, and five percentage points of ten thousand is five hundred. So the quit line produces about five hundred extra quitters a year.

**Kiffer:** Right. And what does an extra five hundred quitters do to the province's smoking prevalence?

**Sarah:** Two hundred thousand minus five hundred is 199,500 smokers out of a million, so prevalence goes from twenty percent to 19.95 percent. The program helps five hundred real people, but you can barely see it in the population figure.

**Kiffer:** That's the gap between how well something works for the people who use it and how many people it reaches. Now the second step. Suppose the province also raises cigarette taxes, and, again with made-up numbers, the price rise leads two percent of smokers to quit. How does that compare?

**Sarah:** Two percent of two hundred thousand is four thousand smokers. That's eight times what the quit line produced, and the tax didn't need anybody to pick up a phone.

**Kiffer:** Exactly. A structural measure acts on every smoker at once, including the ones who would never seek help. And here's where the two ideas combine. Suppose the price rise also prompts twice as many smokers to call the quit line, so ten percent of them call it. What happens?

**Sarah:** There are 196,000 smokers left after the tax. Ten percent of them is 19,600 callers, and five percentage points of that is 980, so roughly a thousand more quitters. Together that's about five thousand fewer smokers, and prevalence goes from twenty percent to about 19.5 percent.

**Kiffer:** Which is the practical message of this question. The individual program works best when the structural change brings more people through its door. I'd judge any behaviour change program by two numbers, its effect on the people it reaches and the share of the population it reaches, and I'd multiply them before drawing conclusions.

**Sarah:** Question three. Christakis and Fowler reported that obesity, smoking cessation and happiness seemed to spread through friendships in the Framingham data. In their obesity paper, which followed 12,067 people from 1971 to 2003, a person's chance of becoming obese rose by fifty-seven percent if a friend became obese. Is behaviour really contagious?

**Kiffer:** The difficulty is that three different processes all leave the same pattern. Influence is when one friend changes the other's behaviour. Homophily is when people who are already alike choose each other as friends. And shared environment is when friends are exposed to the same conditions, so they change together. Let's practise telling them apart with a made-up example. Picture a student residence where friends tend to have similar running habits. I'll give three observations, and you sort each one.

**Sarah:** Go ahead.

**Kiffer:** First, a new trail opens beside the residence, and two friends who live there both start running that month. Second, students who already ran met at a running club and became friends. Third, one friend starts running, and six months later her friend, who had never run, starts too and says it was because she kept being invited along. Which explanation fits each one best? Take a few seconds.

*(Pause)*

**Sarah:** The first is shared environment, because the trail changed the conditions for both of them at the same time. The second is homophily, because the similarity was there before the friendship. And the third looks like influence, because one change came first and there's a plausible route from one person to the other.

**Kiffer:** Good. And notice that the third is only the best guess. The friend might have started running anyway. In survey data, you usually see only that two connected people became similar, and you can't watch why.

**Sarah:** Did Christakis and Fowler try to deal with that?

**Kiffer:** They did. For example, they pointed out that the effect wasn't seen among immediate neighbours, which makes a purely local environmental explanation less likely. But two kinds of critique cut deeper. In 2011, Cosma Shalizi and Andrew Thomas showed that in observational network data, homophily and contagion are generally confounded, so separating them requires strong assumptions. And in 2008, Ethan Cohen-Cole and Jason Fletcher applied similar methods to a large study of American adolescents and found apparent spread of acne, headaches and even height between friends.

**Sarah:** Height. That one's hard to explain as influence.

**Kiffer:** Which was the point. When they adjusted for the conditions that students in the same school shared, those implausible effects shrank, and none of them was statistically significant. So the method could produce contagion where none was possible. The fair conclusion is that friendships matter for health, and that the size of the effect is much harder to pin down than the early headlines suggested.

**Sarah:** Question four takes us to addiction. Insite in Vancouver opened in September 2003, and the evidence that it saves lives is strong. Yet sites keep facing closure. Why hasn't the evidence settled the argument?

**Kiffer:** Let's look first at one piece of the evidence, because it's a good exercise in reading a before-and-after study. In 2011, Brandon Marshall and colleagues published an analysis in The Lancet comparing the period before Insite opened with the two years or so after. Within five hundred metres of the site, the overdose death rate fell from 254 to 165 per hundred thousand person-years.

**Sarah:** So I'd want to know how big a drop that is. 254 minus 165 is eighty-nine, and eighty-nine divided by 254 is about 0.35, so thirty-five percent. Can we say Insite prevented thirty-five percent of overdose deaths nearby?

**Kiffer:** That's the question to ask. What else might have changed over those years?

**Sarah:** Anything that changed across the whole city, like the drug supply or other services. So you'd compare with the rest of Vancouver.

**Kiffer:** Right, and in the rest of the city the rate went from 7.6 to 6.9 per hundred thousand person-years, a drop of about nine percent that wasn't statistically significant. The much larger fall close to the site is what makes the result persuasive. It's still observational, though, since a site like this can't ethically be randomized.

**Sarah:** And the Supreme Court of Canada relied on this kind of evidence in 2011, when it ordered the federal health minister to grant Insite's exemption.

**Kiffer:** It did. Yet the politics have kept moving. British Columbia's decriminalization pilot was narrowed in 2024, and the province let it end on January 31, 2026. The reason the evidence doesn't close the debate is that the disagreement is partly about values. Some opponents accept that sites prevent deaths and still object to any official accommodation of drug use, or worry about disorder near a site.

**Sarah:** So the evidence answers the question of whether sites save lives, and the public argument is often about something else.

**Kiffer:** That's how I'd put it. Evidence settles empirical disputes. When the dispute includes values, the most a researcher can do is make the trade-offs clear, including the deaths that follow when a service closes.

**Sarah:** Question five, and this is the one where Kiffer and I disagree. Vivek Murthy's 2023 Surgeon General's advisory said that lacking social connection can raise the risk of early death to levels comparable to smoking fifteen cigarettes a day. I have real doubts about that comparison.

**Kiffer:** Then let's start with what the underlying studies found, so we're arguing about the same thing. In 2015, Julianne Holt-Lunstad and colleagues pooled cohort studies and found that social isolation was linked with about a twenty-nine percent higher likelihood of death, loneliness with about twenty-six percent, and living alone with about thirty-two percent, in studies that adjusted for several confounders.

**Sarah:** Here's a calculation I'd like everyone to try, because a relative number like twenty-nine percent means little until you put a baseline under it. Imagine, and these are made-up numbers, a group of older adults with strong social ties, ten percent of whom die over ten years. Roughly what would you expect in a similar group that's socially isolated? Take a few seconds.

*(Pause)*

**Kiffer:** As a rough guide, ten percent times 1.29 is about thirteen percent. Strictly, those studies report odds ratios, and converting properly from a ten percent baseline gives about twelve and a half percent. Either way, it's two or three extra deaths for every hundred people.

**Sarah:** Which is meaningful. My objection is to the cigarette comparison. Smoking has decades of evidence, a dose-response, clear biology, and natural experiments. The isolation estimates come from observational cohorts, where people who are ill often become isolated because they're ill. Putting the two side by side borrows the certainty of the smoking evidence for a much less certain estimate.

**Kiffer:** That's a fair point about reverse causation, and it's the strongest version of your argument. Where I'd push back is that the comparison is about the size of the association, and on that measure it's reasonable. An earlier meta-analysis of 148 studies found that people with stronger social relationships had about fifty percent higher odds of survival, and isolation also predicts heart disease, depression and cognitive decline. I think the comparison has helped get a neglected exposure onto policy agendas, which counts for something.

**Sarah:** But a vivid comparison can also push governments toward quick individual fixes, like a befriending app, when the drivers are housing, transit and work.

**Kiffer:** I agree with that risk. That's why I'd use the comparison to open a conversation about structural conditions, and I'd say clearly that it's an association from observational studies.

**Sarah:** I'm still more skeptical than Kiffer, but we agree on the practical rule. If you use the fifteen cigarettes comparison, name the source, say that it describes a similar-sized association, and put an absolute number beside it.

**Kiffer:** Let's pull it together with three things to take away.

**Sarah:** First, when an exposure is very common, compare the people who escaped it. Doll and Hill's percentages looked close, and the odds showed a fourteen-fold difference.

**Kiffer:** Second, judge any individual program by its effect on the people it reaches times the share of people it reaches. Structural measures act on everyone at once, and they can also send more people to the individual programs.

**Sarah:** And third, ask what else could produce the same pattern. Homophily and shared environment can look like contagion, a citywide trend can look like a local effect, and illness can produce isolation as well as the reverse.

**Kiffer:** If you'd like more practice, rework the quit line problem from question two with a reach of two percent of smokers, and see how much a tax would need to do to match it.

**Sarah:** Next time, it's Lesson nine, Environmental Health: Indoor, Built, and Natural Environments, where we look at how the places we live and work shape our health.

**Kiffer:** Thanks for working through some heavy topics and some careful numbers with us. Take care, everyone.

**Sarah:** See you in Lesson nine.
