How do Scientists (particularly social scientists) know which variables to control for, and which not to?
Suppose I for some reason had the hypothesis that people with lower empathy made better businesspeople and I conducted some research and controlled for socio-economic background. If my hypothesis was correct but low empathy was also heritable and this meant lower empathy people were more likely to have higher socio-economic backgrounds then I might find no relationship. Isn’t this a problem, ie controlling for a variable meant I got the wrong result? What am I missing?
In: 2
When you try to do a study of this nature, you’re only likely to find correlation and not causality. The problem is not likely to be “control of variable” but poor formulation of hypothesis and/or insufficiently available data to support the research. At least, that would be my take.
When you try to do a study of this nature, you’re only likely to find correlation and not causality. The problem is not likely to be “control of variable” but poor formulation of hypothesis and/or insufficiently available data to support the research. At least, that would be my take.
When you try to do a study of this nature, you’re only likely to find correlation and not causality. The problem is not likely to be “control of variable” but poor formulation of hypothesis and/or insufficiently available data to support the research. At least, that would be my take.
[Relevant xkcd](https://imgs.xkcd.com/comics/confounding_variables.png).
When you don’t control for enough/the right variables, your data might lead you to a bunch of correlations without causation. If you control for too many variables, you risk shaping the data yourself or removing the source of a true effect.
The answer is therefore that there is no clear answer, and that in the case of social sciences some problems just can’t be conclusively answered with an observational study design like you describe.
[Relevant xkcd](https://imgs.xkcd.com/comics/confounding_variables.png).
When you don’t control for enough/the right variables, your data might lead you to a bunch of correlations without causation. If you control for too many variables, you risk shaping the data yourself or removing the source of a true effect.
The answer is therefore that there is no clear answer, and that in the case of social sciences some problems just can’t be conclusively answered with an observational study design like you describe.
[Relevant xkcd](https://imgs.xkcd.com/comics/confounding_variables.png).
When you don’t control for enough/the right variables, your data might lead you to a bunch of correlations without causation. If you control for too many variables, you risk shaping the data yourself or removing the source of a true effect.
The answer is therefore that there is no clear answer, and that in the case of social sciences some problems just can’t be conclusively answered with an observational study design like you describe.
It can but generally it can also be accounted for. Let’s take your example here “people with low empathy are better businesspeople.”
You could take a collection of random people, test them for business proficiency, test their empathy levels, and then sort them by socioeconomic background. If your hypothesis is true that a lack of empathy makes someone a better businessperson, then you should note that trend regardless of socioeconomic background.
And this may be important because you might find a compounding variable in these results. Perhaps people in low socioeconomic conditions are better businesspeople with high empathy, and the proposed trend only appears in high socioeconomic conditions. Or you might find evidence of the contrary, that being low empathy makes you better at business, being better at business makes you low empathy.
It can but generally it can also be accounted for. Let’s take your example here “people with low empathy are better businesspeople.”
You could take a collection of random people, test them for business proficiency, test their empathy levels, and then sort them by socioeconomic background. If your hypothesis is true that a lack of empathy makes someone a better businessperson, then you should note that trend regardless of socioeconomic background.
And this may be important because you might find a compounding variable in these results. Perhaps people in low socioeconomic conditions are better businesspeople with high empathy, and the proposed trend only appears in high socioeconomic conditions. Or you might find evidence of the contrary, that being low empathy makes you better at business, being better at business makes you low empathy.
It can but generally it can also be accounted for. Let’s take your example here “people with low empathy are better businesspeople.”
You could take a collection of random people, test them for business proficiency, test their empathy levels, and then sort them by socioeconomic background. If your hypothesis is true that a lack of empathy makes someone a better businessperson, then you should note that trend regardless of socioeconomic background.
And this may be important because you might find a compounding variable in these results. Perhaps people in low socioeconomic conditions are better businesspeople with high empathy, and the proposed trend only appears in high socioeconomic conditions. Or you might find evidence of the contrary, that being low empathy makes you better at business, being better at business makes you low empathy.
Ideally you control by running multiple experiments each testing a different hypothesis. However funding is a limiting factor so often times you just do the best you can with what you have af hope someone else reads what you did and can fill in the gaps with their own research.
Latest Answers