1
$\begingroup$

I'm looking to demonstrate an association between penalty (overtime) payments and absenteeism. The dependent variable is overtime payments ($0-$highval) for a shift, and the outcome variable is 'was the person absent (sick) for this shift'.

What I'd planned on doing was: a) bucketing overtime payments eg. $0, $0-$50, $51-$100 etc. then b) correlating this against '% of shifts with this particular penalty payment where the person 'went sick' (which is a variable that is either 'at work' or 'absent'.

So the output would be something like:

Penalty      %Shifts not worked
0              13%
50+            6%
100+           3%

This obviously requires a bit of data pre-processing, which I'd like to avoid in taking a first-pass look at the issue.

Is there a more basic/efficient way to test a continuous dependent variable like $ overtime paid and a 0-1 variable (at work/absent)? Thanks guys Pete

Improve this question
$\endgroup$
3
  • $\begingroup$ Have you considered doing a simple regression on the problem? I would give more suggestions but I am unsure by your problem statement if your outcome is continuous or a 0-1 variable? $\endgroup$
    – user25658
    Sep 22, 2013 at 6:42
  • $\begingroup$ BabakP, the outcome is a 0-1 variable: they person was either sick or at work. $\endgroup$
    – Pete855217
    Sep 22, 2013 at 8:39
  • $\begingroup$ Ok, then I believe you mean to say independent and not dependent here: "The dependent variable is overtime payments..." because the word dependent is usually associated with the outcome variable. $\endgroup$
    – user25658
    Sep 22, 2013 at 15:37

1 Answer 1

Reset to default
1
$\begingroup$

So I understand you have a binary variable and a continuous variable. Unless you have a very good reason why, I don't suggest categorizing the overtime payments variable as you are simply throwing away information into more simpler and less useful forms

You can simply use a pearson correlation between the two. This is called point-biserial correlation. The p-value you get with the correlation is the same p-value you get with a t-test. You can also calculate the confidence interval of the mean difference in pay between the categories at work and absent.

More importantly, before you calculate any numbers, I hope you plotted the data. Use a side-by-side boxplot, one for at work, another for absent. This should be more telling than any test. For example, if you identify extreme outliers, non-parametric inferences may be better.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thanks Hotaka and BabakP - I will simply run a regression through r and do the boxplot (whisker too). Hotaka, yes I did have a concern that I was throwing out data by categorising. In terms of the PHP/MySQL code, I've also taken a quick look by calculating % sick leave shifts for each category of penalty (there are only 5 - 0, 15%, 17.5%, 22.5% and 100%). By ignoring the dollars (which are continuous) and focussing on the rates only (categorical) I'm able to simplify my MySQL/PHP processing, which was the main concern with trying to identify the right approach. I'll use R to do the rest. Tks $\endgroup$
    – Pete855217
    Sep 22, 2013 at 8:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.