Employee Performance Project Help

JDC

New Member
Hello all,

I have a project I'm working on and I need some advice. I have only a very basic knowledge of stats. Here goes...

Your company has a production department with three shifts and a total of 115 employees. There are four factors used to measure an employee's performance in this department - Quality (QA), Productivity (PROD), Attendance (ATT), Level of Work Difficulty (WkDIFF). You've been asked to find a way to use the data from the four aforementioned factors to create a single number that will assign an overall performance rating for each worker. The Quality measure is similar to a test score (points remaining out of possible 100), the Productivity measure is a ratio (Amount work completed : Hours worked), the Attendance measure is amount of points remaining (out of nine, employees are terminated upon accruing nine attendance pts), and the Work Difficulty is an average of work difficulty ratings assigned to certain tasks. Hopefully that makes sense.

The only way I know to fairly combine these scores (that are in different units of measurement) is to convert them to standard scores then sum each employee's scores for each variable which would give the final rating number.

QA + PROD + ATT + WkDIFF = Performance Rating.

Here's my questions:

1) The rating will be calculated for each shift and overall for the entire department, should each shift be treated as a sample of the population (entire department) or should each shift be treated as it's own separate population?

2) I'm not performing an experiment, just trying to give employees a fair and accurate rating of their overall performance. What other statistical analysis can be used to ensure the data is accurate and reliable.

3) How should I examine the distribution of the data - for each individual variable or for the final performance rating scores? What are the possible pitfalls if the data is not normally distributed?

Sorry to ask so many questions. I just need to know that my thinking on this calculation isn't flawed. I also just need ideas on how to find ways to make that data tell me more than just the performance rating numbers (if that's possible).

The other problem, I have to be able to figure all this out using only Excel.

Any suggestions or help is GREATLY appreciated.

By the way, nice website...I'm lucky to have found it.

JDC

JohnM

TS Contributor
JDC,

There are literally hundreds of ways this can be sliced and diced - what is termed as "fair" is debatable...

-in the real world, companies will often work something up based on economics/market conditions and human resource policies rather than a rigorous statistical analysis (i.e., what pay increases are we willing to give out based on employee performance).

....and an employee's final performance rating is often not a "hard" calculation - it's often a simple categorized rating like "meets requirements" or "exceeds requirements" that is also subject to management's judgment.

It basically boils down to an ordinal or categorical rating system, with the top 10-20% getting a certain raise, the middle 80% getting a certain raise, and the bottom 10% not getting a raise.

The reasons for this are that a "hard" calculation, especially when it involves money and people's livelihoods, is subject to scrutiny and "legal" pitfalls. Additionally, people in management like to keep things very simple.

However, if this is merely a classroom project/exercise for a business or human-resource management curriculum, then I'm guessing that your teacher has some expectations around the quantitative methods you've been studying......

Often when a composite score is made up of the sum of component scores, and the component scores are in very different scales:

QA (100)
Productiivty (a ratio, so it theoretically can be anywhere between 0 and infinity)
ATT (9)
Work Diff (an average of something - scale undefined here)

the method involves translating raw scores into standard scores, as you suggested. Excel should be able to handle that quite easily.

Assuming that each component is equally weighted, then the overall score should be the sum of the components.

If the distributions aren't normally distributed, then there are obvious pitfalls - some people wil be given ratings they don't deserve - either too high, or too low.

JDC

New Member
Employee Performance Project

Thanks for the reply, I appreciate the insight.

So how would you go about calculating statistics, in regards to treating each shift as it's own population or just a sample of the entire department?

Also, could you suggest maybe a few specific tests/measurements that would (as my professor likes to say) "make the data talk."

Does Excel have a function (besides graphs) that will allow me to test for normal distribution?

Thanks again for you help.

JDC

JohnM

TS Contributor
So how would you go about calculating statistics, in regards to treating each shift as it's own population or just a sample of the entire department?

I would not treat each shift as its own population - if the shift's average is low, then people on that shift who are "above average" may not be "above average" when compared to other shifts - it's just a basic fairness / equal expectations thing.

Also, could you suggest maybe a few specific tests/measurements that would (as my professor likes to say) "make the data talk."

Can't really help you here. You'll need to do some exploratory work to see if you notice any unusual or interesting patterns.

Does Excel have a function (besides graphs) that will allow me to test for normal distribution?

No, it does not.

JDC

New Member
Many thanks, you've been quite helpful.

Hopefully I won't have to bother you too much more.

JohnM

TS Contributor
You're definitely not bothering us.

If you haven't noticed, we kinda like this stuff - otherwise we wouldn't be participating.....

JDC

New Member
Cool...

I'm struggling with the basic stats of this project (if can't already tell). My planned course of action is to collect data for a while then as you suggested, do some exploratory examination, then once I've got a larger amount of data I'll decide on alternate measurements and/or tests depending on any patterns that arise. I'm thinking that examining correlation between variables is a good place to start. My biggest problem now is finding a way to test the distribution for normality having only limited software capability. A lot of the stats I'd like to run depend on having normally distributed data.

I appreciate the fact that you take the time to offer free advice, I can't thank you enough.

pmjimenez

New Member
Hi JDC, sorry for the late response....I think what you're looking for is the ANOVA testing. This allows you to test more than two groups' means. You would use each group's scores individuall for this as well as in between groups. Each department will have a mean and deviation. Then you will use the total mean of all groups to calculate the rest...I'm quite new to this so I may be wrong, but it sounds like you need to do ANOVA testing. You can find the ANOVA experiment testing on most stats websites or here is a link that explains it pretty well: http://faculty.vassar.edu/lowry/webtext.html
Also, the more scores you have, the more normal the distribution...
Hope this helps...

JohnM

TS Contributor
pmjimenez,

Actually ANOVA won't really do anything for you in this situation - JDC isn't trying to see if there are differences between/among groups - yet, anyway.

He was just trying to get a fair way of determining each employee's performance rating.

JohnM