I am currently wiriting on my master thesis and have a major problem concerning my OLS regression. The results are contrary to the scientific literature and I think it is because the variables have no lineare relation. Maybe someone knows how to transform the following graphs to get a more linear relationship.
Take out outliers and keep influential data points? Unless you crop up the y axis, your data does not seem to have much correlation/relation. Calculate the correlation coefficient and you can see how strong that relationship is. I can assume you have the raw data right?
Henry Green
how about sounding less like a sperg u tryhard loser lmoa :^)
idk what "the literature" in your field is but you could like not use linear regression, make a log transform or be better at collecting data, pick one
Nicholas Thompson
I cannot take the outliers out and yes I have the raw data. I tried log and exp but did not work out either. I am only familiar with ols and logit, but I have no clue for further processing
Robert Thompson
echoing this sentiment. your data seems to have different populations within it.
Adam Bennett
yes they are 3 different products. the problem is only if the linar relation is strong enough and I dont think so. Sadly, I cannot figur out a transformation. But thanks to any advice from you guys
Dylan Gomez
>The results are contrary to the scientific literature That's good enough user, economy is not a science after all.
Carson Scott
OP you'd probably perform better at whatever it is you're doing if you could SPEAK FUCKING ENGLISH
Landon Turner
that´s probably true :-)
Blake Harris
For the first two, try recoding both variables as their logs, sometimes that can make nonlinear relationships linear
As for the last one, I'm not sure how you can fix that shit
Nolan Flores
Where exactly is the problem? If the data doesn't fit the previous literature, it just doesn't.
Present your findings, say that previous research has different conclusions, talk about your methodological limitations (sampling problems, bias etc.). There is no problem if you manage your results right.
That being said, your second graph looks like it has a curve-like relationship. Try nonlinear regression models, might help.
Connor Torres
try using a linear mixed model that has an estimator incorporating shrinkage
Matthew Lee
what's your end goal? Is it to illustrate the relationship between these two variables or to produce a graph that looks like a figure in a textbook on linear regression?
Hunter Sanchez
those labels make me angry. you can fit a linear model to EVERYTHING the RMS-Error can be huge though. did you expect the "relative price difference" to rise in your attributes? if so your data might just be shit or different to the data in the "literature".
also try taking absolute values of relative differences
Leo Robinson
stop using OLS
Landon Lopez
or if you insist on using least squares, transform the X's so they have a linear relationship with the Y. People have said heteroskedasticity might be happening, however if you are only willing to use OLS you can't really do anything about that. If you do want to tackle the potential heteroskedasticity you'll have to generate wieghts by regressing your residuals against your predictors, and use those weights to re-estimate the regression (google weighted least squares).
Dylan Bailey
This guy right here. This is the guy.
Andrew Morgan
This might be a good resource for correcting the issue with your data. I've used this SPSS code before.
If all you're concerned about is whether there is a relationship and aren't specifically concerned about specific parameter estimates you could try ranking data and then doing regression.
Leo Scott
>SPSS
Nolan Fisher
I've actually learned how to do extensive coding in SPSS