Hello,
We're conducting quantitative research as a part of this semester, and the analysis method of choice is multivariate linear regression. Right now, we're checking for the fulfillment of assumptions, however, the scatter plots to check for linearity are bit off. We use number of patents as our dependent variable, but differences between some of the countries vary too much. So the scatter plot looks like this. Could you please recommend us how to go about optimizing the dataset, without removing too many observations from the data set? The most extreme outlier on the top is China, which we removed already, however that does not fully solve the problem. I was thinking of creating groups within the DV with specific intervals, and coding it 1, 2, 3, 4, etc..
Dear Mr Korec, just a few remarks or call it questions or suggestions... wouldn't it be possible to group countries based on a number of variables like gdp or national research efforts...etc. Secondly many patents are qualitatively not so strong. How do you deal with that issue? Third, did you use the inventor's nationality? Some countries enfavour patent systems for a number of age categories, like cheaper filing fees for aged inventors. You'll have to take this in your analysis...etc. All the best.
Hello Filip,
In addition to the "apples to apples" concerns raised by Nico Deconinck , there's one other obvious concern to note from your query.
It certainly looks as if, were you to stick with ordinary least squares regression (univariate or multivariate), you might wish to transform the DV of patents (perhaps log transform). Since this is a counting variable, you might think about whether to use poisson regression or negative binomial regression instead.
Good luck with your work.
I think you need adjust the scale on the Y axis to get a better sense of how X & Y are related. Because the current version is set to represent the unusually high value of a single outlier, that collapses the distribution of the vast majority of the cases. So, I recommend removing that outlier and then re-running scatter plot.
One reason for doing this is that your current relationship does look like a zero-slope straight line, but the range on Y is so restricted that you cannot really tell.
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking