We are a “clicking” nation. We measure how important ideas are by the veracity of readers’ “like” “share” or comments. In response to this reality, newspapers’ headlines are dramatic and “clicking good”. Becoming increasingly common however, is that such click-baiting headlines rarely reflect the content of the news items. Human foibles promote the constant search for dramatic new insight, in contrast to the slow dribble from the fountain of knowledge.
A case in point is a paper published in the esteemed medical journal Lancet this year. The news reports went ecstatic with this news, within a few days reports spread across the globe with 402 newspaper and magazine headlines reporting: “Million Women Study: Happiness does not guarantee a long life” (Times UK-Dec 10, 2015); “Happiness is not the key to a long life” (Independent Online-Dec 10, 2015); “Happiness Doesn't Bring Good Health, Study Finds”(New York Times-Dec 9, 2015); “Happiness Doesn't Help You Live Longer” (The Atlantic-Dec 9, 2015); “Being happy doesn't make you live longer, research says” (The Independent-Dec 9, 2015); “Is Happiness Really Linked To Longevity? Maybe Not, Study Finds” (Forbes-Dec 10, 2015). In one short article, and a contagion of newspaper headlines, a century of work that has consistently shown the importance of happiness for longevity have been ignored or discarded. We “clicked” happiness to death. But this might have been exaggerated.
The original paper [1], conducted by Bette Liu with the Faculty of Medicine, University of New South Wales, Australia and her colleagues, had an interesting methodology. The short abstract—on which most newspapers must have based their compelling news reports—concluded: “In middle-aged women, poor health can cause unhappiness. After allowing for this association and adjusting for potential confounders, happiness and related measures of wellbeing do not appear to have any direct effect on mortality.” Being miserable is cool and it does not hurt you. But before you throw away your Prozac, lets examine what these researchers did and see if we can come up with different and more articulate interpretation of the data.
This study was based on the Million Women Study, a study of 1.3 million UK women in their 50s who went for a mammogram—recruited between 1996 and 2001—looking at future cancer incidence after taking Hormone Replacement Therapy. In the past, the results based on this database have been controversial. The overall findings have found lifestyles, habits or behaviors as remembered and recalled by participants in their youth have less an affect on cancers, heart disease or other illnesses than present lifestyle choices. Basically arguing that—other than radiation—the present is more important than the past. This goes against the whole field of epigenetic studies and how lifestyle and diet choices have long lasting effect, sometimes across multiple generations. But one of the problems with this database is that they are dealing with recall and self-perception. How we recall events tend to come in-line with our present reality.
Returning to this specific article on happiness, heralding the importance of happiness, surprisingly the authors found that in all the results, being happy most of the time faired better on ALL variables. No exceptions. While being unhappy was associated with higher mortality—even after adjusting for age, which influences probability of death—by 30 to 40%. Despite these dramatic correlations, the authors still conclude that: “After allowing for this association and adjusting for potential confounders, happiness and related measures of wellbeing do not appear to have any direct effect on mortality”. The fact that the authors were able to whittle away all of the very positive correlates of happiness to leave a shriveled edifice of happiness, holding no predictive power, attest to how skilled the authors are at playing with statistics. Within a clinical database, the authors undress happiness until its emptiness is exposed. But what is happiness without its expression?
It is not the statistics that is questionable but the authors’ methodology. This type of analysis is referred to as “kitchen sink” analyses. Chuck everything in and see what comes out. Results are not theory driven but motivated by spurious and random associations. Eliminating correlates of happiness away—within a very limited, clinical database—happiness becomes irrelevant. But happiness is an collection of evaluations of how content we are in life. It is made up of individual components, with our evaluation of our health forming a main aspect of our happiness.
Because the authors found that happiness was related to all positive variables, they did something very strange. They adjusted happiness. Within logistic regression—which tests the effect of a condition/s (or independent variable) on a yes or no outcome (dependent variable, which in this case was dead or alive) you adjust variables by modifying the outcome to match the condition and thereby eliminating the effect of one independent variable on the dependent variable. This is important because you can isolate an individual variable and see how it behaves regardless of all other variables. Which is what the authors did for happiness. First they broke up the groups into three main groups and then they started eliminating variables sequentially to see which one will mute the effect of happiness. They did this by diluting their construct validity, and then by reverse engineering happiness. Let me explain.
The first methodological fault is massaging their definition of “unhappy”. Out of 719,671 women with a median age 59 year, 39% reported being happy most of the time while the majority (44%) reported being happy usually, while the final group was defined as rarely reporting happiness. But this is not accurate. This final group was composed of three very distinct categories of people who reported being either happy sometimes, rarely or never. Far from being a homogenous group this category is a subjective potpourri of a group of sensible people who report that they do sometimes feel happy, combined with—according to the DSM-V—a clinically diagnosable group who are likely depressed and report never to be happy. So this is a strange mixture of people grouped together and called “unhappy.” Methodologically, the authors should have selected only the never feel happy group. They are a distinct group. But by lumping all three categories together they lost construct validity. We do not know what they are comparing the happy group against. When they refer to the “unhappy” group a proportion of these are happy sometimes.
They continued to dilute the construct validity by excluded the first five years of follow-up and women who had already had heart disease, stroke, lung disease, or cancer. We do not know why these people were excluded but it is likely that these very ill women were likely to be the least happy and the most clinically depressed (based on their own data of who these women are because the strongest associations with reported unhappiness were treatment for depression or anxiety and reporting only fair or poor general health). By eliminating them the authors got rid of the negative outliers, further diluting the construct to include people closer to the average. By including “sometimes happy” people in the group of “unhappy” people, and then eliminating the extremely unhappy people, what the authors did is to dilute the construct of “unhappy” to produce a group that is closer to the average.
The second methodological fault is the reverse engineering of happiness. After finding that happiness is positively correlated with all healthy indicators, the authors proceeded to strip away these variables. This is known as adjusting the data. The authors adjusted the data for a number of factors. In regression analyses such adjustments create ceteris paribus, a Latin term meaning “all else being equal.” So when you adjust for variables you even out—eliminate the affect of—that variable. Practically you are throwing these variables out of the effect. Such statistical techniques are important when you want to see if one variable is important by itself accounting for the effect of all the other variables separately. But in this case we have to question the number of variables that were adjusted to minimize happiness. Theoretically in psychology, happiness is not a stand-alone construct but an omnibus construct reflecting a number of individual components. If you eliminating these recorded expressions—and the Million Women Study database is limited in how happiness is recorded—then there are few variables that are correlated. Being happy was correlated with increasing age, having fewer educational qualifications, doing strenuous exercise, not smoking, living with a partner, and participating in religious and other group activities.
Only when the authors eliminated ALL the correlated of happiness that exist in their database did happiness become an emaciated variable with no meaning. The authors first adjusted only for age. Then they continued to adjust for region of residence at recruitment including employment, car ownership, home ownership, and household overcrowding, college and pre-college education, living with a partner, whether they are obese, perform strenuous exercise, smoke, drink alcohol of one drink a day, participation in religious or other group activity. None of these activities diminished the effect of happiness, which tells us that happiness is expressed in people who are not defined by any of these categories. The only variable that seems to mimic or act as a proxy for happiness is self-rated health (in their Table 2).
In summary the three adjustments that got rid of the happiness factor:
Eliminate all the really depressed and sick people. Eliminating over 125 769 women who at baseline already had heart disease, stroke, cancer, or chronic obstructive airways disease. These excluded women who had three times the death rate. Again, it is very likely that this group of women were the most depressed and unhappy.
Diminish the effect of older adults who are normally happy. By adjusting for age we are reducing the effect of happiness. We know that the older we become the happier we are. Such consistency data that has economists interested in psychology because it determines economic behavior. Despite this adjusting for age, happiness still emerged as a resilient factor in reducing mortality. Adjusting only for age, unhappiness remained associated with 25-33% increase in death.
Only by getting rid of self-rated health did the effect of happiness completely vanished. Translated this means if we eliminate the importance of how healthy or unhealthy participants felt, then it does not matter how miserable you are in determining your mortality.
Diminish the effect of older adults who are normally happy. By adjusting for age we are reducing the effect of happiness. We know that the older we become the happier we are. Such consistency data that has economists interested in psychology because it determines economic behavior. Despite this adjusting for age, happiness still emerged as a resilient factor in reducing mortality. Adjusting only for age, unhappiness remained associated with 25-33% increase in death.
Only by getting rid of self-rated health did the effect of happiness completely vanished. Translated this means if we eliminate the importance of how healthy or unhealthy participants felt, then it does not matter how miserable you are in determining your mortality.
But after adjustment for self-rated health, treatment for hypertension, diabetes, asthma, arthritis, depression, or anxiety, lifestyle factors—including smoking, deprivation, and body-mass index— unhappiness was not associated with mortality from all stress or lack of control.
In psychology, happiness is relatively stable, while unhappiness is more variable [2]. Similarly the authors of this study reported that there was some instability in happiness figures, especially from being unhappy to becoming happy a year later. While only 2% who reported being happy most of the time at baseline changed to being unhappy at follow-up, 5% of women who reported being unhappy at baseline reported being happy most of the time a year later. This is a gain of 3 % per year (difference between becoming happier to becoming sad). From their own study the results show that very year, there is an improvement on happiness of 3%.
Dariusz Leszczynski, a Polish cell biologist wrote in the Washington Times Communities, Oct 3, 2013 that “The Million Women Study has shoddy exposure design leading to shoddy results and ending with shoddy conclusions.” Applying a database to study relationships other than what the database was original developed for is not inherently bad science. But when there are complex constructs such as happiness, that are not fully understood, having a million or more women who went in for a mammogram might not be a representative group to generalize from. The limitation in external validity is significant.
Happiness is a central emotional indicator that brings our body and mind in balance. It is one of the main predictors of mortality that even economists and actuaries apply to adjust their mortality forecasts based on the present level of happiness and self-rated health. If there is one objective in life it is to be happy, everything else is peripheral. Attempting to summarily dismiss a century of research [3]—that has been trying to understand the meaning of happiness and longevity—needs to be questioned.
Citations
[1] Liu, B., Floud, S., Pirie, K., Green, J., Peto, R., Beral, V., & Million Women Study Collaborators. (2015). Does happiness itself directly affect mortality? The prospective UK Million Women Study. The Lancet.
[2] Veenhoven, R. (1994). Is happiness a trait?. Social indicators research,32(2), 101-160.
[3] Lucas, R. E. (2007). Personality and the pursuit of happiness. Social and Personality Psychology Compass, 1(1), 168-182.
© USA Copyrighted 2015 Mario D. Garrett
No comments:
Post a Comment