Which Hypothesis Test Should I Use?

Six Sigma – iSixSigma Forums General Forums Tools & Templates Which Hypothesis Test Should I Use?

This topic contains 14 replies, has 5 voices, and was last updated by  Mike Carnell 6 months, 1 week ago.

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
  • #56014


    Quick question. In my manufacturing facility, we got the paint job on our units being affected by craters that we are suspecting might be caused by a specific brand of operator’s gloves.

    So I’ve got four sets of data, one pair with attribute and one pair with variable data.

    For both of these sets I got the number of defective/defects (depending if its attribute or variable data) within the sample when there we were using the suspicious gloves and also the same info when a new brand was introduced.

    Here’s what my raw data looks like in case I’m not doing a good job describing itHere's what my raw data looks like in case I'm not doing a good job describing it

    Which hypothesis testing do you guys recommend is more convenient to see if there is significant impact of these gloves?


    Robert Butler

    The picture of your data is too small to read – how about posting it as an excel attachment so we can see what you have?

    Based on your description and by doing a lot of squinting it looks like you have recorded a defect count along with the total number of things checked over on the left hand side and the same thing on the right hand side and all you have done is label one attribute and one variable. Assuming I am reading your picture correctly, in both cases if you divide the defect count by total number of items examined you will have a percentage and you can run a two sample t-test on the percentages with the population identifier being glove type.


    Prasenjit Guru


    First of all Number of Defect is not variable data, its count data…

    For comparing two proportion you can do 2 Proportion Test and For Comparing Count data you can do two sample Poisson rate test…


    Robert Behrens

    Previous poster is correct. Data is not true variable data, but it is quasi-variabledata due to the fact you developed it using correlated variables. You could use a 2-sample T with the quasi-variable data. You can use test for 2-proportions for the ratio of defect created with each set of gloves. One other method you could use is to measure the surface area of all defects on the parts, sum them to generate actual variable data. Another method I have used is to Zone the part, count the zones with a defect and develop a percentage of coverage that can be used with 2-proportions.


    Robert Butler

    “Variable are classified as continuous or discrete, according to the number of values they can take. Actual measurements of all variables occurs in a discrete manner, due to precision limitations in measuring instruments. The continuous-discrete classification, in practice, distinguishes between variables that take lots of values and variables that take few values. For instance, statisticians often treat discrete interval variables having a large number of values (such as test scores) as continuous, using them in methods for continuous responses.”
    – From Categorical Data Analysis 2nd Edition – Agresti pp.3

    Based on what I can read from the OP’s picture of his spread sheet he has a sufficient spread in counts which means there shouldn’t be any problem analyzing the data using the methods I outlined in my initial post.


    Marc Thys

    The hypothesis test question has been answered in the previous posts, however I would like to comment from another perspective.
    From ogling the data, even if the difference between gloves may prove to be significant, it does not look like it will help you much in improving the process. Both cases appear to have similar levels of defects / defectives. In other words, any difference may be significant but not relevant / important!
    This is something I keep on emphasising in projects such as these: do not confuse significance with importance.
    Secondly, if you don’t have a clear explanation of WHY the gloves may be yielding different results, it does not help much with the actual root cause analysis.
    However, the data you have could be helpful in generating other hypotheses. It would seem that especially the defective rates are jumping up and down. You could set up a control chart (p chart) to detect if there are “unusual” (ie special cause) events in the data. Then try to correlate those events with other things that happened at the same time.
    In any case, it would appear you need a much deeper understanding of the process and the potential causes for defects before digging deeper.


    Stan Alekman

    You might identify a difference between the gloves that you consider to be of practical or useful significance (value) in choosing between the gloves. With this difference in mind, you can do a test for superiority.



    First of all, sorry for the reply, this forums will just not notify me.

    So this takes place at a manufacturing facility for bbq kettlegrills ([like this one](

    Once the grills are assembled, they go through an inspection line where two processes happen:
    1.- Inspection: grills are checked by operators for any defects (in this case, craters on paint) and if any found, they input them on a computer system. This is where the variable data comes from in the sheet im sending you (number of defects on total units)

    2.- Repair: Operators check their computers on each grill and if they find any defects they look for them in the grill and either repair them (good unit) or send them back to repaint as a rerun (defective units) This is where the attribute data comes from in the sheet im sending you (defectives from a sample)

    Before this inspection line, there’s a few prep, finishing and assembly lines, where operators wear gloves to protect themselves and the units when manipulating them. We suspect these gloves might be causing craters.

    So as you see in the data, from Feb 20th to March 26th operators were using Vargas Brand. And from March 27th up until June 9th, operators used Ansell Brand.

    I want to see if there’s statistical evidence to prove gloves did or didnt have and impact on craters.

    [Raw Data](

    Thank you so much again for doing this. Im only allowed to use Minitab 17 on my computer so if you can work in that, would be awesome, if not, its cool. Let me know of any questions.


    Marc Thys

    I ran the t-tests and the differences are significant.
    However, as I pointed out in my post, that does not mean the gloves are the main cause for craters – the data actually suggest otherwise.
    Vargas gloves overall get poorer results, but they occasionally are better.
    And, occasionally, Ansell gloves are worse!
    Also, bear in mind that switching gloves may have had a psychological impact, and that operators have been paying more attention since they got the new gloves.
    Bottom line: the glove brands may have an impact, but switching won’t solve the cratering problem.
    I think the control charts are interesting. Have a look in the file that I attached.


    Chris Seider

    Did you see anything with the graphical/statistical tools yet?


    Marc Thys

    Here is a google drive link to the file:
    usp=sharing”>Gloves MTB 17



    Thank you for your reply Marcy. I think the link got cut down I cant see any attachments. Is there a way you can send it via email? javiero [at] outlook [dot] com




    How the hell do I edit a post? My bad Marc! I misspelled your name. Thank you for your reply @marcthys I think the link got cut down I cant see any attachments. Is there a way you can send it via email? javiero [at] outlook [dot] com



    Mike Carnell

    @javiero Regardless of the type of hypothesis test your glove issue is not going to eliminate the issue 100%. I am assuming that on a finish surface no defects are ok. You might consider some analysis that opens up more opportunities to identify the issues. One previous post suggested mapping the defects which is a good idea. You also may want to consider some type analysis like Multi-Vari and start narrowing this down.

    Paint is a process that is sensitive to a lot of variables. You might want to look at how these items are cleaned prior to paint. Surface condition is a big factor in finish quality. Just as an example there is a factory where they paint cars. The painters are not allowed to wear deodorant because they went though a real finish issue and ended up chasing it down to a particular deodorant a person was wearing. Deciding this is a glove issue seems a little premature. I would put some more effort into defining the problem.

    Just my opinion.


    Marc Thys

    Trying to post the MTB file again:
    Gloves MTB

    Hope it works this time.

Viewing 15 posts - 1 through 15 (of 15 total)

You must be logged in to reply to this topic.