Monday, November 9, 2009

Science Lesson: Parenting Style Study

A BBC article discussing a study of parenting was a part of my standard morning blog and news feed*.  The article said that a 'tough love' parenting style, combining warmth and discipline, was linked to positive character traits such as self-regulation and empathy.  I was intrigued by the study on parental style and life outcomes it talked about, so I decided to take some time to dig deeper.

It took a little bit of searching to find the article they were referencing.  This, by the way, is one reason why bloggers are superior to the old-school journalists who just don't understand how to operate in the Internet age.  No self-respecting blogger would ever talk about a research study without providing a convenient link to it. 

Note: What follows may seem like a long ramble, but I think it is a useful running commentary of how scientists analyze things.

After I opened the report, the first thing I did was look for an abstract.  I didn't find one, so I started reading the introduction.  But after a few sentences of pointless babble and overblown prose, I realized that it would be a waste of my time.  So I jumped straight to the technical appendix and started analyzing the methods of the study and the raw results.

The first thing I noted was that the outcome variable of the survey was a behavior analysis of the children at age five.  The text of the BBC article alluded to this but did not make it explicit.  The first sentence, however, was wrong.  It said "Children brought up according to "tough love" principles are more successful in life, according to a study."  But the study did not actually tell us anything about success in life.  It simply connected parenting style to childhood character traits that are believed to lead to later success in life.

The second thing that jumped out was that the outcome variables were not based on independent observation.  They were based on a survey of the parents.  So in the end all the survey can actually proves is that there is a connection between how parents treat their kids and how they view their kids' behavior. This could be a potentially fatal flaw, but to their credit the authors acknowledged it as a weakness.  That admission of weakness, by the way, is a strong signal of good science.

Their statistical treatment of income seemed odd to me.  Instead of listing it as a continuous variable, which would have told us useful things about the effects of increasing wealth, they only used it to isolate the poorest 20% and put them in a different category.  All they can say about income is the differences between that category and the rest.  Their treatment of the data prevents them from saying anything useful about income effects in the upper 80% of the sample.  This method seems primitive to an economist, and raises suspicions of data mining, but it may be the standard way of doing things in sociology.

I like the way they split parenting style into two dimensions: rule enforcement and attachment.  'Tough love' parents, the ones that produce the best outcomes, are those who combine high attachment and high rule enforcement.  They are both warm and controlling.  Second place is 'laissez-faire' parents are warm and permissive.  Third place are 'authoritarian' parents who are hostile and controlling.  The worse results come from 'disengaged' parents who are hostile and permissive.

But they only asked two questions for the 'rule enforcement' scale, and each one only had three possible responses.  This means there are only six possible values for a key independent variable.  That is probably too coarse of a scale to capture the effects well.  I am fairly sure that the state of the art in social science surveys is to ask for responses on a seven-point scale from 'strongly agree' to 'strongly disagree' and make scales from at least four of five questions each.

At this point, I had to refer back to the text of the paper to see details of the childhood character traits they were measuring.  One thing that I noticed is that five of the ten of the measures of 'empathy and attachment' are, at least partially, measures of what other people are doing to the child.  If a child is bullied or rejected by other children, that child is listed as having low empathy.  So a child will score poorly on the outcome scale if he or she suffers from discrimination that is not his or her fault.  Given how vicious and tribal little kids can be, and how easily they absorb prejudices from their parents, I do not belive this is a fair assessment.  If this survey is taken in a society that has any discrimination at all, it will imply that minority parents do worse at raising well-adjusted children.

There is a similar, but smaller, problem with the 'self-regulation' scale.  If a child has any medical problems that lower quality of life, he or she will be listed as having worse self-regulation because of the way the questions are written.

But, despite these problems, these scales are probably still good at predicting life outcomes.  People who score low on these measures will probably have more troubles in life, even if those troubles are not their fault.  So the scales are accurate as predictors, but it would be wrong to assign moral weight to them, or to say that a parent whose child scores badly has dome a bad job.

But the biggest flaw in that paper is that it simply shows correlation, not causation.  It is not a controlled experiment.  It might be the case that both parental style and child behavior are determined by genetics, or something in the environment.  It is less likely, but still possible, that intrinsic attributes in the child cause parents to react differently.  The only way to really test these effects would be to randomly tell caregivers to have more or less attachment and rule enforcement for different children.  This is both impossible and unethical. 

But it might be possible to find some kind of 'natural experiment' that causes variation.  Studying adoptive parents would be a good start.  The closest you could get to a controlled experiment would be to send identical twins to two different sets of adoptive parents, who live in equally good neighborhoods and have the same amount of money and status, and who are equally competent, but have different philosophies about raising children.  Then you could be very confident that any differences were caused by parenting styles.  But doing that, or finding children that match those conditions, would be really hard, so we do what we can.

It may seem strange to say this after I have criticized the study so much, but it is a good piece of research.  Yes, they made mistakes, but science is a messy process, especially when combined with surveys and political initiatives.  The core of the paper is good, and it makes a useful contribution to human knowledge.  It tells us things about the link between child-rearing actions of parents and perceptions of attributes that impact future life success.  It allows us to replace hazy anecdotes and folk wisdom with real data.

The key result from the paper, which is probably true even with all of my caveats, is that the most important thing to do for children is to love them.  Attachment is key.  After that, it is important to have clear rules and enforce them consistently.

Some people might read my blog post and say 'I already knew all that.  I don't need some fancy scientist to tell me what any good grandmother knows about kids.'  And it is true that many people already do know these things.  But many people don't.  Some would argue that being authoritarian is best.  This study suggests that they are wrong. 

Even if it seems silly, you have to use science to test conventional wisdom.  Often, the conventional wisdom is wrong.  And when different groups of people have different sets of conventional wisdom, only science can tell who is right.

Another important lesson from this is that scientific knowledge is often indefinite and probabilistic, and should often be interpreted narrowly.  This study does not deliver certainty.  The knowledge it presents is a somewhat muddled link between what parents do and how they see their kid's behavior at age five.  But that narrowness gives it strength.  Its flaws could be identified, and fixed by a better survey.  More data will come in about the life outcomes of the children.  We will improve our knowledge.  This survey, flawed though it is, will still be useful as a data point in 100 years.

By contrast, people who 'just know' things about the world will claim to have all the answers and be very certain about their knowledge.  They might write books offering sweeping claims without any proof.  But the whole thing falls apart under pressure.  They don't have any good reasons for believing things.  Their 'knowledge' is often nothing more than superstition, or the current fashion of the culture they are in.  In a generation, it will all look hopelessly outdated, and will be revealed to be worthless.

It is always useful to read claims about the world made long ago.  They tend to look amazingly primitive and stupid.  But it is important to remember that people believed them.  Often the people who write them had doctorate degrees, and everyone believed that their words were the words of 'science'.

People are often admonished: 'Don't accept anything without proof.'  But this statement is useless.  People almost always think that they have good proofs for everything they believe.  The key is to learn what is and is not proof.  The words of a scientist, or any other authority figure, are not proof.  The fact that lots of people believe something is not proof.  The fact that people have done something for hundreds of years and it seems to work is not proof.  Proof can only come from carefully analyzing the results of a controlled experiment, or something we observe that is a lot like a controlled experiment.

* I basically make up my own custom online newspaper by combining dozens of different news sources and commentators.

No comments: