Regression Analysis

Question

I have the salary dataset that contains skills as single column(Independent Variable) and Salary as Dependent Variable. Then I split the skill column into multiple Skill column based on its presence 0 or absence 1. eg: emp_id skills Salary 1 R,python,excel,word 4000

I made dataset transition like this:

emp_id R Python Excel word Java Salary 1 1 1 1 1 0 4000

Then i performed multiple linear regression, to find out the skills influencing salary most. I have summary of results.

My question is that, is the only analysis we can do or what are all the other alternative analysis we can do to predict the salary.

Answer 1

I understand that you use a continous variable (salary) as an outcome which you may want to predict by a categorical variable (skill). Because one individual may have multiple skills (a Word user may also have Excel as a skill, but not necessarily R or Python) you perform multiple linear regression.

Depending on your dataset you may first want to consider whether you have enough observations in each skill group to back up your analysis and make adequate predictions using this model.

You may want to consider that fitting a prediction model for salary based on these computer skills, but no other information that would explain more of the salary variation (e.g. education status, years of employment, previous job level), may be inadequate to actually predict the salary of individuals with other backgrounds. I would call this a threat towards the generalizability of your results.

If you don't want to predict the salary but are interested in differences overall, an alternative may be to directly compare salaries between defined groups (eg R vs Python; or either of the two vs Word/Excel only). This may be achieved, depending on the number of observations per skill and the number of factors you want to compare, by simple t-test or ANOVA accompanied with a few boxplots.

Answer 2

Do you care about interpretability of your results, or just about the prediction accuracy? If accuracy is what you are looking for, you could try using other machine learning algorithms (Boosted decision trees for instance).