Sample Size When comparing two different groups..
Six Sigma – iSixSigma › Forums › Old Forums › General › Sample Size When comparing two different groups..
 This topic has 8 replies, 4 voices, and was last updated 18 years, 3 months ago by R.

AuthorPosts

June 3, 2004 at 7:52 pm #35748
I need some help, I am surveying a group of customers that call into our call centers, and need to know how to get a representative sample. The problem I am having is how to determine this since there are two different call centers and the call ratio is about 4:3 on the volume for each center. In a baseline study both had essentially the same mean and std. dev for their satisfaction score….could someone point me to how to determine the sample size of all calls that will get me proper representation across both call centers??? Thanks in advance.
0June 3, 2004 at 9:33 pm #101208Peter,
Can you give a litle more detail on what you’re trying to accomplish?0June 4, 2004 at 12:42 am #101212
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.Here are some questions:
1. You say they had the same mean and s.d.. Did you treat sat scores as continuous or discrete data? How r u measuring cust sat?
2. What level of confidence do you want to achieve?
3. What level of precision do you want to achieve?
4. What estimation can you make about the population variation?
5. Do you want to sample separately from each or from the combined population?
6. What is your research question? What do you want to learn from the data?
Once you provide the answers to these questions, the answer will become clear.0June 4, 2004 at 1:59 am #1012141. Customer sat is a discrete number, on a scale of 19. The means were A) mean=5.04, st dev 3.01 b) mean=4.94, st dev 2.992. 95%3. +/ 5%4. Not sure what variation you are talking about5. want to sample from a combined population. It is easier for us to link to back to the call center after the results after the fact.6. We want to be able to do site level comparisons between the two call centers are we roll out changes.Thanks for all your help. What I originally did was used a sample size calculator I had found out on UCLA’s stat’s site, it took the mean of each population, the combined standard deviation (which was something like .11 for these), power = .80 and confidence interval of .05. It told me I need to sample 32…that cannot be right….hence my reason for posting here.Thanks for all your help!
0June 5, 2004 at 1:43 am #101269
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.A couple of thoughts:
1. You say the Cust Sat is a discrete number. Was the mean and standard deviation calculated as discrete values or continuous? I am getting a sense that you used continuous calculations. That would account for the seemingly small sample size.
2. One of the variables in the sample size calculations is an estimate of the population variation.
3. Sampling from the combined population might best be accomplished through a stratified random sampling strategy to account for any significant differences in population size.0June 5, 2004 at 1:11 pm #101281
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.Some additional thoughts. I assume that Cust. Sat. is gathered through an end of call phone survey. Have you determined response rate and considered the implications of self selection and non response bias? What is the BIG Y in your survey? Have you correlated the survey questions with the BIG Y? Have you considered not worrying about the entire scale and just do “top 2 box” or “top 2 box minus bottom 2 box”? Are your scores normally distributed or severely skewed? If so, is mean the best measure or is median? Do you do Sat. with Organization, Sat. w/ Call or Sat. with Associate? Or all of them. Are calls randomly selected?
0June 7, 2004 at 4:14 pm #101338Darth, thanks for all the questions, felt like i was in a firing line!
The Cust. Sat is gathered through random outbound calling so there is no nonresponse bias in that sense. The big Y is Overall Satisfaction. The scores are skewed, a lot of 1’s and 9’s. We do sat with many things, overall, rep, org, etc…they are randomly selected.
Okay, i think I am really confused on why taking the mean of the scores is not good and what I should do instead. If possible, will you provide me your email address, I think I need a lot more help then I previously thought.
Thanks!0June 7, 2004 at 4:52 pm #101339
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.June 24, 2004 at 8:08 am #102259Peter,
My understanding is that u require to know how to sample calls and not the size of sample.
If my understanding is right, please segment first on Customer satisfaction since that is the Big Y that you are addressing. You would then need to take samples of the customers appearing in the segmented population.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.