Column Jeroen Mulder
Statistically speaking
During a statistics lecture on establishing cause-effect relationships, my lecturer came up with an example. It was in a 2010 article by Robert E. Larzelere, an American professor of Human Development & Family Science. He compared different methods for correcting problem behaviour in children: which approach can parents best use to reduce the unsocial or antisocial behaviour of their child? Professional interventions, such as using Ritalin or visiting a therapist? Or are parents better off addressing problem behaviour themselves, by sending the child to their room, yelling, or physically punishing them?
The lecturer showed a table of results. Four different statistical techniques were used to analyse the collected data: two techniques commonly used in sociology and econometrics, and two techniques that have a long tradition in psychological research. The same research question, the same data, but four different calculations to arrive at an answer.
The teacher started talking more and more enthusiastically. As it turned out, different analyses showed different results. Take snapping and shouting at your child, for example. According to one analysis technique, there was a positive correlation (more shouting and snapping leads to more problem behaviour), according to a second technique there was no correlation, and the last two techniques resulted in a negative correlation (more shouting and snapping leads to less problem behaviour). Professional interventions supposedly had no effect or actually led to more problem behaviour, depending on which analysis technique you used. And hitting your child could do no harm, according to two techniques.
The take-home message was clear: the interpretation of research results can depend on the analysis technique chosen, and the right technique is not always obvious. Especially in complex phenomena, such as children’s behaviour and development, statisticians and methodologists have an important role. Together with researchers, they look at what is the best design for a study, and what statistical techniques are needed to reach valid conclusions. Statisticians are also continuing to develop these analysis techniques so that researchers can answer new types of questions.
At the end of the lecture, I asked the lecturer which correction method she personally preferred. She left that question unanswered.
Want to know more? In the episode Statistiek met Jeroen Mulder from the Dutch podcast Makkelijk Praten, Jeroen explains why statistics are so difficult and interesting at the same time.
Jeroen Mulder (1994) is a PhD candidate in statistics at the CID.
This article is part of a New Scientist special issue about the Consortium on Individual Development, that will appear in September 2023.
