Return to Answer

Regression towards the mean is a far more common problem than is often realised.

It is also one of those things that is actually quite simple but appears to be quite nebulous on closer inspection, and this is partly due to the narrow way that it is usually taught. Soemtimes it is attributed entirely to measurement error, and that can be quite misleading. It is often "defined" in terms of extreme events - for example, if a variable is sampled and an extreme value observed, the next measurement tends to be less extreme. But this is also misleading because it implies that it is the same variable being measured. Not only may RTM arise where the subsequent measures are on different variables, but it may arise for measures that are not even repeated measures on the same subject. For example some people recognise RTM from the original "discovery" by Galton who realised that the children of tall parents also tend to be tall but less tall than their parents, while children of short parents also tend to be short but less short than their parents.

Fundamentally, RTM is a consequence of imperfect correlation between two variables. Hence, the question shouldn't be about when RTM occurs - it should be about when RTM doesn't occur. Often the impact may be small but sometimes it can lead to copletely spurious conclusions. A very simple one is the observation of a "placebo effect" in clinical trials. Another more subtle one, but potentially much more damaging is the inference of " growth trajectories" in lifecourse studies where conditioning on the outcome has implicitly taken place.

Regression towards the mean is a far more common problem than is often realised.

It is also one of those things that is actually quite simple but appears to be quite nebulous on closer inspection, and this is partly due to the narrow way that it is usually taught. Sometimes it is attributed entirely to measurement error, and that can be quite misleading. It is often "defined" in terms of extreme events - for example, if a variable is sampled and an extreme value observed, the next measurement tends to be less extreme. But this is also misleading because it implies that it is the same variable being measured. Not only may RTM arise where the subsequent measures are on different variables, but it may arise for measures that are not even repeated measures on the same subject. For example some people recognise RTM from the original "discovery" by Galton who realised that the children of tall parents also tend to be tall but less tall than their parents, while children of short parents also tend to be short but less short than their parents.

Fundamentally, RTM is a consequence of imperfect correlation between two variables. Hence, the question shouldn't be about when RTM occurs - it should be about when RTM doesn't occur. Often the impact may be small but sometimes it can lead to completely spurious conclusions. A very simple one is the observation of a "placebo effect" in clinical trials. Another more subtle one, but potentially much more damaging is the inference of " growth trajectories" in lifecourse studies where conditioning on the outcome has implicitly taken place.

Regression towards the mean is a far more common problem than is often realised.

It is also one of those things that is actually quite simple but appears to be quite nebulous on closer inspection, and this is partly due to the narrow way that it is usually taught. Soemtimes it is attributed entirely to measurement error, and that can be quite misleading. It is often "defined" in terms of extreme events - for example, if a variable is sampled and an extreme value observed, the next measurement tends to be less extreme. But this is also misleading because it implies that it is the same variable being measured. RTM may arise fo measures that are not even repeated measures on the same subject. For example some people recognise RTM from the original "discovery" by Galton who realised that the children of tall parents also tend to be tall but less tall than their parents, while children of short parents also tend to be short but less short than their parents.

Fundamentally, RTM is a consequence of imperfect correlation between two variables. Hence, the question shouldn't be about when RTM occurs - it should be about when RTM doesn't occur. Often the impact may be small but sometimes it can lead to copletely spurious conclusions. A very simple one is the observation of a "placebo effect" in clinical trials. Another more subtle one, but potentially much more damaging is the inference of " growth trajectories" in lifecourse studies where conditioning on the outcome has implicitly taken place.

Regression towards the mean is a far more common problem than is often realised.

It is also one of those things that is actually quite simple but appears to be quite nebulous on closer inspection, and this is partly due to the narrow way that it is usually taught. Soemtimes it is attributed entirely to measurement error, and that can be quite misleading. It is often "defined" in terms of extreme events - for example, if a variable is sampled and an extreme value observed, the next measurement tends to be less extreme. But this is also misleading because it implies that it is the same variable being measured. Not only may RTM arise where the subsequent measures are on different variables, but it may arise for measures that are not even repeated measures on the same subject. For example some people recognise RTM from the original "discovery" by Galton who realised that the children of tall parents also tend to be tall but less tall than their parents, while children of short parents also tend to be short but less short than their parents.

Fundamentally, RTM is a consequence of imperfect correlation between two variables. Hence, the question shouldn't be about when RTM occurs - it should be about when RTM doesn't occur. Often the impact may be small but sometimes it can lead to copletely spurious conclusions. A very simple one is the observation of a "placebo effect" in clinical trials. Another more subtle one, but potentially much more damaging is the inference of " growth trajectories" in lifecourse studies where conditioning on the outcome has implicitly taken place.

That regression towards the mean is a far more common problem than often realised.

Regression towards the mean is a far more common problem than is often realised.

It is also one of those things that is actually quite simple but appears to be quite nebulous on closer inspection, and this is partly due to the narrow way that it is usually taught. Soemtimes it is attributed entirely to measurement error, and that can be quite misleading. It is often "defined" in terms of extreme events - for example, if a variable is sampled and an extreme value observed, the next measurement tends to be less extreme. But this is also misleading because it implies that it is the same variable being measured. RTM may arise fo measures that are not even repeated measures on the same subject. For example some people recognise RTM from the original "discovery" by Galton who realised that the children of tall parents also tend to be tall but less tall than their parents, while children of short parents also tend to be short but less short than their parents.

Fundamentally, RTM is a consequence of imperfect correlation between two variables. Hence, the question shouldn't be about when RTM occurs - it should be about when RTM doesn't occur. Often the impact may be small but sometimes it can lead to copletely spurious conclusions. A very simple one is the observation of a "placebo effect" in clinical trials. Another more subtle one, but potentially much more damaging is the inference of " growth trajectories" in lifecourse studies where conditioning on the outcome has implicitly taken place.