Home / General / Maybe You Can Fearlessly Stay Grumpy. Or Maybe Not.

Maybe You Can Fearlessly Stay Grumpy. Or Maybe Not.


You may have already read about the Lancet study (Liu et al.) casting doubt on the claim that happiness is related to longevity. The study had a sample of 720,000 middle-aged (50-69) women from the UK Million Women Study, recruited through breast cancer screening. They were asked a single question about how happy they were (as part of a large battery). Then they were followed for 10 years for multiple health outcomes; in this study, the outcome under investigation is mortality. Liu and her co-authors found that after controlling for lifestyle factors that you would expect to be related to happiness and also be health promoting or damaging (like smoking), and especially after controlling for self-reported health, there was no relationship between happiness and longevity. They conclude that there’s no strong evidence that happiness affects longevity; it may just be that poor health affects both happiness and longevity, so you see a statistical relationship between happiness and longevity even though there’s no direct causal link. Participants were also asked questions about feeling stressed, relaxed, and in control, and the study similarly finds no relationship between longevity and all these survey questions.

Ed Diener, Sarah D. Pressman and Sonja Lyubomirsky, happiness researchers with some skin in the game, have three main critiques of the Liu paper. The first is that the happiness measure was flawed: it was a single question near the end of a battery of more than 300; if the researchers wanted to study happiness they should have given participants an opportunity to rate the frequency they felt a variety of positive emotions, and when surveys are that arduous, individual item validity is called into question. Since stress, they say, is non-controversially associated with health outcomes, if they didn’t find a relationship between stress and longevity maybe their measures aren’t valid.

The second is that, given that the women were already middle-aged when data collection started, their happiness could have already affected their self-reported health:

The average participant was 59 when she enrolled in the Million Women Study. The women self-reported their health (from “poor” to “excellent”) at the beginning of the study, and the authors then applied statistics to these data to counter the possibility that the participants’ health itself was what was influencing their happiness, rather than vice versa. This makes sense, but it changes the nature of the question being answered. After all, a woman’s happiness for the first six decades of her life could have affected her health, yet the analysis essentially took that out of the picture, as though the middle-aged subjects were a blank slate. The authors’ question, then, was “Does happiness at 59 predict life span beyond its effects during the first six decades?”

Their final critique is that Liu and co-authors control for too many lifestyle factors — sleep, exercise, smoking, etc., etc. — even though these may be the pathways by which happiness effects health. I find their last criticism the least compelling. The Lancet authors’ whole stated aim is to test whether there is robust evidence for a direct effect of happiness on longevity after accounting both for the relationship between health and happiness and a bunch of lifestyle factors associated with being happy. They are doing the analysis that answers that question. But it’s important not to over interpret that finding. Just as an analysis that shows that the the effect of race on an outcome is reduced or disappears when you control for socio-economic status (SES) does not prove that there is no direct effect of race on that outcome, the fact that the effect of happiness on longevity disappears when you control for a host of factors does not prove that there is no direct effect of happiness on longevity. To “control for a variable” in a regression analysis is to not let it vary. If, for example, you are looking at the effect of happiness on longevity, but you control for smoking, in your final calculation of the strength of the effect of happiness, you don’t include the variation in happiness that is correlated with smoking. But that doesn’t actually mean that happiness isn’t having a direct effect. It just weakens the evidence that it is. Maybe some of that correlated variation that predicts longevity is due to happiness, and some is due to smoking, and some is due to other factors entirely. A regression analysis cannot tell you.

I wrote two long, heated paragraphs about the Lancet authors’ decision to control for treatment for depression and anxiety, when depression and anxiety are, especially in this context, basically the same thing as being very unhappy, before I read the paper one more time and realized that the effect of happiness disappeared before including mental health treatment in the model. I still think it’s bizarre to control for depression and anxiety in this context— it’s like doing a study on whether donut love predicted how often people went to Krispy Kreme, and controlling for how often participants ate donuts at home — but it’s not affecting their main point. I am glad I stopped myself before my rant fully unspooled.

  • Facebook
  • Twitter
  • Linkedin
This div height required for enabling the sticky sidebar
Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views : Ad Clicks : Ad Views :