Yesterday, the "most viewed" and "most emailed" item on the New York Times website was Deborah Tannen's essay, "Why Sisterly Chats Make People Happier", which opens this way:
"Having a Sister Makes You Happier": that was the headline on a recent article about a study finding that adolescents who have a sister are less likely to report such feelings as "I am unhappy, sad or depressed" and "I feel like no one loves me."
These findings are no fluke; other studies have come to similar conclusions. But why would having a sister make you happier?
The usual answer — that girls and women are more likely than boys and men to talk about emotions — is somehow unsatisfying, especially to a researcher like me. Much of my work over the years has developed the premise that women's styles of friendship and conversation aren't inherently better than men's, simply different.
Prof. Tannen goes on to give some of the reasons that she finds "the usual answer … somehow unsatisfying", and her thoughts deserve the attention that NYT readers gave them. But I was curious about the power of sisterhood, whether it might come from "styles of friendship and conversation" or from something else, and so I took a look at the recent study that she links to: Laura Padilla-Walker et al., "Self-regulation as a mediator between sibling relationship quality and early adolescents' positive and negative outcome", Journal of Family Psychology, 24(4): 419-238, 2010.
There's a lot to like about this study. For example, the authors worked with a large sample that was not limited to undergrads in a psych department's subject pool — an on-going longitudinal study of nearly a thousand children in a demographically balanced set of nearly 400 families.
The dependent variable that they used to measure "(un)happiness" was self-report of what they call "internalizing behaviors":
Externalizing and internalizing behaviors were assessed at Time 1 and 2 from a measure taken from the Youth and Family project assessing antisocial behavior (9 items) and depression/anxiety (13 items). Sample items for externalizing include, "I lie or cheat" and "I steal things from other places than home;" and for internalizing include, "I am unhappy, sad, or depressed" and "I feel that no one loves me." Participants answered these items in regard to the adolescent's behavior on a scale ranging from 0 (not true) to 2 (very true or often true).
Unfortunately, the authors give us only a highly digested form of the results. We don't get the raw tables of independent and dependent variable values — a row for each child, with demographic variables including sibling(s) sex along with that child's measure of "externalizing and internalizing behaviors". We don't even get mean values and standard deviations for the "internalizing behaviors" measure among sisterful and sisterless children (unless I managed to read past them). Instead, we get the "betas" (standardized regression coefficients) and associated significance levels from an SPSS AMOS "path analysis":
So in order to give you a more concrete idea of what the beta=-0.13 (connecting Sibling Gender to Internalizing Behavior in two-parent families) and beta=-0.12 (for the same connection in single-parent families) might mean, I created some fake data (by running the obvious regression equation as a generator). A histogram from a typical run looks like this:
For this particular set of fake data, the standardized regression coefficient for predicting "internalizing score" from presence of absence of a sister is -0.128, significantly different from 0 at the p<.05 level, just as in the authors' analysis of the real data. The mean internalizing score for the 500 "children" with sisters is 13.11, and the mean score for the 500 "children" without sisters is 12.60. (Recall that these scores are the result of summing the answers to 13 questions like "I feel that no one loves me" on a three-point scale, "from 0 (not true) to 2 (very true or often true)". So -0.51 — the average effect of having a sister — might be interpreted as meaning that half of the sisterful respondents lowered their response to one of the 13 questions by one unit.)
Pearson's r for the correlation between sisterlessness and unhappiness is 0.0638, so the proportion of variance (in "internalizing score") accounted for by having a sister is r^2 = .004.
(The difference in mean values in the true experimental data might have been larger or smaller than this, and the actual percent of variance accounted for might have also been larger or smaller. I's not possible to tell from the information given in the paper, as far as I can see, though I'd be surprised if the truth were radically different from this simulation, which represents the results of my first guess about what some fake data with the observed normalized regression coefficients might look like.)
These fake results, like the real results, are statistically significant — but how meaningful are they?
The answer depends on why you want to know. If you want to know why you're so (un)happy, then not very much of the answer can be expected to depend on whether or not you have any sisters. One way to express this in an intuitive way is to observe that if we randomly pick from this (fake) distribution a child with sister(s) and a child without, the sisterful child will have a lower "internalizing" score — report themselves as less depressed — about 53% of the time. 47% of the time the sisterless child will be happier on this measure.
This is a long way from the statement that "Adolescents with sisters feel less lonely, unloved, guilty, self-conscious and fearful", which is how ABC News characterized the study's findings, or "Statistical analyses showed that having a sister protected adolescents from feeling lonely, unloved, guilty, self-conscious and fearful", which is what the BYU press release said. At least, if you take predications about generic plurals ("A's are X-er than B's") to be statements about typical members of the groups involved, then such statements are false. On the other hand, such statements are true if you take "A's are X-er than B's" to mean simply that a statistical analysis showed that the mean value of a sample of X's was higher than the mean value of a sample of Y's, by an amount that was unlikely to be the result of sampling error.
Unfortunately, I think that most people interpret general statements about generic plurals as claims about typical members of the groups involved — even when the sentences are hedged with statistics-talk like "tend to" or "are less likely to". In this particular case, the results of misunderstanding are probably harmless or even beneficial — no doubt thousands of NYT readers were motivated to have sisterly chats, and even some of the sisterless may have thought to look around for an honorary sister or two, which is surely all to the good. The fact that the misinterpretation reinforces a commonly-held stereotype is part of the reason for the enthusiastic uptake. But not all commonly-held stereotypes — and not all generic plural statements by scientists — are so benign.