When a stud says they controlled for all other variables, how do they do this?

166 views

[ad_1]

I understand what “controlling for” means. The concept is not the issue. It is more about what the actual process is. It seems so powerful and so simple. I must be missing something.

In: Mathematics
[ad_2]

You mean a _study_ yea?

It just means that they have looked at some other potential causes of the effect they are seeing and have concluded that they wouldn’t explain it. For example, if you are studying which fertilizer works best, you might compare yields from farms that have used different fertilizers, but you’ll want to check that those farms are otherwise similar – if it turns out that the low-yield farms all had low rainfall, that might be a better explanation. Ideally your analysis would take into account differences in rainfall by, for example, only comparing farms that had similar rainfall.

Clearly it’s never possible to rule out *all* other potential explanations. In my experience this is one area where it’s particularly common for researchers (and certainly secondary sources such as press releases and media articles) to exaggerate the findings slightly by being overly dismissive of alternative explanations. I’m not sure there is really ever a context where it would be appropriate to use the phrase “we’ve controlled for all other variables”.

There are two ways that they do this. One way is by forcing all the other possible variables to be the same across both halves of the study. This is the preferred method since everything is literally the same other than the one variable that you’re changing and investigating its effect on the result.

Another way is to mathematically correct for the variables that are different across the two halves of the study. For example, if one half of the study was performed at 80 Celsius and the other half at 70 Celsius, and you know that temperature increases your result by 1% per Celsius, then you can reduce your 80 Celsius half’s result by 10% to simulate what that half of the experiment would have done if it were performed at 70 Celsius.

The problem with doing things this way is that you rarely have a perfect correction. In addition, multiple variables being different means you have to do multiple corrections, but sometimes those multiple variables interact in complicated ways so you can’t just add or multiply the corrections together.