# Why linear regression is widely adopted given its (sometimes) poor prediction

Linear regressions are in nearly every paper I've read. The authors always model the response variable as a linear combination of independent variables without regards to how the data is actually generated. It seems like they only use linear regression because it can produce a table with CI of "influence" of each variable ($$\beta$$ in $$y = \alpha + \beta x + \epsilon_i$$).

I want to know reasons social scientists use linear regression this much.

Also, should I use a more opinionated model (with a better fit) if possible?

• Often linear regression is used but the axis of the graph are scaled to turn non linear data into linear data. So you can get a non-linear function if you rescale the axis back to their original but it's nice because it's easy to calculate correlation coefficient and R-squared. – ajax2112 Mar 23 '20 at 8:40

## 1 Answer

As someone with background in social science I would say two main reasons:

Ignorance; usually social science get a crash course in statistics with simpler tests such as, chi-square, t-test, linear/logistic regression and ANOVA. This limits the way of looking at at problem. If all the tools you have are hammers, every problem will look like a nail.

Good enough; In social science its seldom that you want to predict a response variable rather you want to explain it. in that case a linear regression will do good enough. since there is no need for "a perfect fit". Sure if you want to maximize R-squared/RMSE there are other solutions, but if you "only" want to get an understanding what drives a behaviour, a general idea from a linear regression is often good enough.

I hope this was an answer to your question.