Success and Self-Control
Joshua Wiley takes us through a published study that monitors long term outcomes relating to measurements of childhood self-control. He explains how the study was done and the results interpreted. Note that the published study is #openaccess , so do check it out!
Originally shared by Joshua Wiley
Relations of Childhood Self-Control to Health and Success
This is not a new article in PNAS, but it details results from a unique study and I have wanted to write about it for a while. NOTE: this is a 1,200 word post.
The punch line is the authors find self-control in childhood predicts a variety of important outcomes decades later.
Rather than start with the cool results, I am going to start with the methods to consider the question, “What are the data and how strong is the design to support the results?”
The article reports data from the Dunedin Multidisciplinary Health and Development Study (see the study website here: http://dunedinstudy.otago.ac.nz/). The study follows a sample of over 1,000 New Zealanders born in 1972-1973 and assessed starting at age 3, and followed up at 11 additional times all the way until they were 38 (the study is ongoing and will follow up again at 44/45).
Multi-decade longitudinal studies are rare, and the design offers the ability to make strong predictive claims. At the age of 3, many deleterious health behaviors (e.g., smoking, drug use, low physical activity) have yet to emerge. Thus it is possible to assess predictors of outcomes with relatively little contamination. Many longitudinal studies are too late or too short to make this claim. For example, assessing whether self-control at age 18 y predicts drug use at age 19 y, it may well be that drug use at age 16, caused a deterioration in self-control at age 18 y as well as higher likelihood of drug use at age 19 y. Thus measuring the predictor before the outcome is not sufficient. Studies assessing predictors decades apart from outcomes render such bidirectional influences less likely (although still not impossible).
What is meant by “self-control”
As important as the timing and length of data collection in a study is how variables were measured. This may be particularly tricky in psychology where variables measured are not directly observable. For example, there is little ambiguity if a study reports a relationship between height and … However, what about a study reporting a relationship in children 11 – 14 y between cheating and impulsivity? Does “cheating” mean the same thing if it is objective instances where a child was caught cheating versus self-report?
In the present article, childhood self-control was created by transforming 9 variables to have the same range and adding them all up. Specifically,
1) observer rated lack of control (e.g., “low frustration tolerance”, “restless”, “impulsive”, “requires attention”) at ages 3 and 5
2) parent and teacher reports of impulsiveness and aggression (e.g., “flies of the handle”) at ages 5, 7, 9, and 11
3) parent and teacher reports of hyperactivity (e.g., “has short attention span”,) at ages 5, 7, 9, and 11
4) Parent and teacher reports of additional hyperactivity items (e.g., “difficulty sitting still”) at ages 9 and 11
5) Parent and teacher reports of lack of persistence (“fails to finish tasks”, “easily distracted”) at ages 9 and 11
6) Parent and teacher reports of impulsivity (e.g., “has difficulty awaiting turn”) at ages 9 and 11
7) Self-report hyperactivity (e.g., “restless”) at age 11
8) Self-report inattention (e.g., “difficulty paying attention”) at age 11
9) Self-report of impulsivity (e.g., “talking while others are still talking”) at age 11
So when the study refers to “self-control” what is really meant is not the theoretical definition, but observer-, parent-, teacher-, and self-report on those items. As a side note, this information is buried in the SI Appendix.
For better or worse, some items are included that might not match common definitions of self-control (here quoted from google: “the ability to control oneself, in particular one’s emotions and desires or the expression of them in one’s behavior, especially in difficult situations”) with multiple items assessing hyperactivity, such as being restless. One could even question whether “fails to finish tasks” is appropriate — what if the child did not desire to finish the task? Is that lack of self-control, or just a difference between what tasks parents/teachers think the child should finish and what the child wants to finish?
What are the outcomes?
There are too many outcomes to go through in detail (but to find out about all their measurement, see the SI Appendix). I will focus on adult (age 32 y) physical health. 892 of the study members provided blood samples. Physical health included 5 categories:
1) Metabolic: weight, blood pressure, HDL and LDL cholesterol, HbA1c, and VO2max were assessed, each was categorized as “at risk” or not, and then these 0/1 risk scores were summed. Having 3+ risk factors was considered clustering of metabolic abnormalities
2) Respiratory function was using a spirometer
3) Periodontal disease (gingival recession, probing depth per tooth in all four quadrants)
4) Sexually transmitted infection (herpes simplex virus type 2)
5) Inflammation measured using C-reactive protein
Often studies reporting on physical health, report self-report measures (e.g., “I am more/less healthy than most people my age”). A strength of this study is detailed, objective, and broad characterization of multiple measures of physical health.
Other outcomes included: _substance dependence_ assessed using both a structured interview as well as informant ratings; depression using a structured interview; adult socioeconomic index based on classifying participants occupations; income based on self-report categories; single-parent child-rearing whether/when participants had children and who the children lived with; financial planfullness based on self-report attitudes toward savings and reports of things such as owning a home, investments, retirement plan; financial struggles based on self-report difficulties meeting the cost of various necessities and other non necessary activities (e.g., holidays) and self-report credit problems; informant-rated financial problems based on informant reports of two items: whether the participant was a poor money manager or lacks enough money to make ends meet; criminal convictions based on search police database of participants from the ages off 17 to 32 (note that this means criminal convictions are not really just at 32 so are much closer to the ages self-control was assessed).
What are the Cool Results?
A very nice table summary is shown here: http://www.pnas.org/content/108/7/2693/T1.expansion.html
Basically, low childhood (age 5-11) self-control was statistically significantly** associated with adult (mostly age 32 but see notes on outcomes above):
–worse physical health
–higher substance dependence (self-report as well as informant reported problems)
–higher likelihood of single-parent child rearing
–less financial planfullness
–more financial struggles (both self-report and informant reported)
–higher likelihood of criminal convictions
All of these results held up when family SES and child IQ were controlled, to try to statistically answer the question, “Does self-control matter if all the children came from families of the same SES and all children had the same IQ?”
Considering family SES, childhood IQ, and childhood self-control, self-control tended to have larger effects on the outcomes, with a few exceptions. Childhood IQ and family SES were more strongly related to SES and Income than was self-control, unsurprisingly. However, self-control was a better predictor of financial planfulness and struggles than even family SES or IQ.
Family SES and childhood IQ were stronger predictors of physical health than was self-control. Low family SES was predictive of worse adulthood physical health, as was low childhood IQ. Self-control also had a statistically significant effect, but it was weaker than SES and IQ.
Where can I find out more?
Read the full text here: http://www.pnas.org/content/108/7/2693.full Pleasantly, the article is freely available without a subscription or having to pay any money for it.
**For non statisticians, “statistically significant” does not mean there is an important or large relationship. Statistically significant indicates that the observed association can be reliably distinguished from zero. With a large enough sample, even very tiny effects that would not be visible or noticeable in daily life can be reliably distinguished statistically.