Comments on: Applied Regression Analysis on Absenteeism and Service Level

By: David

David — Fri, 08 Jun 2012 22:44:03 +0000

In this regression example, did the data come from a population or a (randomly drawn) sample? I cam across this article while researching whether regressions can be performed if you have access to population data. In textbook examples of regression analysis, usually you take a representative sample, run a regression, and make inferences about the unknown population. But, in this example, if your business keeps complete records of service level and employee absenteeism, then would you still need to do random sampling?

In your experience, do business analysts do random sampling?

Thanks,
David

By: Pete Abilla

Pete Abilla — Thu, 05 Jan 2012 18:47:15 +0000

In reply to Rich Neapolitan. Sure.

By: Rich Neapolitan

Rich Neapolitan — Thu, 05 Jan 2012 00:21:59 +0000

I would like to use your linear regression picture in an artificial intelligence text I am writing. Can you grant me permission.
Thanks,
Rich

By: psabilla

psabilla — Mon, 29 Jan 2007 20:09:30 +0000

@Cheddah,

You’re right — the analysis above isn’t perfect by any means. There are 2 specific things I could have done to make the analysis more robust:

(1) Gather more data points — Doing this will ensure that the R^2 is more firm and accurately explains the variation in the data set.

(2) Remove Outliers — Because my data set wasn’t large, the presence of outliers massively influences the direction of the regression line — that is, the outlier draws the regression line towards itself. I can avoid this if I do (1) and also remove outliers where it makes sense to remove.

Given both (1) and (2), I went with the data I did have, warts and all. In the end, the business was able to go with my recommendations and the analysis brought about a positive and pragmatic result. From a researcher’s perspective, I could’ve done much more to make the analysis firmer than it is.

I appreciate the comment and don’t ever worry about being harsh or anything. No big deal at all. Arguments of this type are about ideas, not people. Thank you for your perspective; I appreciate it.

By: Cheddah Yetti

Cheddah Yetti — Fri, 26 Jan 2007 20:40:31 +0000

Since I 1) don’t know you personally and 2) the impersonal nature of the internet makes criticism come off harder than it would in person, I preface this comment by saying I don’t mean to be a jerk with this comment.

However, that analysis is worthless. The influence of the outlier point to the left on the regression line is easy to see from the naked eye. I’m sure you know how to test it, but for those out there not as statistically inclined there is an easy way: remove the point and rerun the regression.

My bet is if you took out the 90%/12% point, the slope now shifts to positive, and r-squared remains high. In all likelihood you’d expect the relationship you found to be true, but if truth was leading this project you’d have to explain the effect that outlier had on the data. Of course, management wouldn’t want to hear it, but it is the truth.

By: Koen lenssen

Koen lenssen — Tue, 05 Sep 2006 14:39:21 +0000

You need more detail no? Otherwise unions could use it in their favour (stress causing sick levels to rise so more staff is needed…).
Interesting analysis though. It’s always good to quantify the information!
BR
Koen