# What Test Do I Use?

Six Sigma – iSixSigma › Forums › General Forums › Tools & Templates › What Test Do I Use?

- This topic has 11 replies, 5 voices, and was last updated 8 years, 2 months ago by Mike Fisher.

- AuthorPosts
- April 3, 2012 at 12:23 pm #54015

JAshouriGuest@jashouri**Include @jashouri in your post and this person will**

be notified via email.Hi There,

I am working a project on returned samples for our soda cans. We send out free samples to large corporations. Some are sent AUTO SHIPPED – and some are sent by our Sales group. I have data that shows the defects for AUTO SHIPPED and by our sales group. A defect is any soda can sample that is returned (Due to various reasons). I wanted to see if there is a statistical difference between defects shipped by AUTO vs SALES. I thought the best test is the CHI SQUARE. IS that correct? Please advise…thanks experts..0April 3, 2012 at 12:44 pm #192942

Chris SeiderParticipant@cseider**Include @cseider in your post and this person will**

be notified via email.Sounds good to me. Just be sure to use # of cans sent defectively and # of cans sent correctly for both categories. As long as you have a minimum count of 5 for each of the four cells in the chart, you can use it.

A more powerful analysis would be to count individually the # of defective for EACH defect type and you can still use chi-squared as long as each cell has at least 5. Just be sure to NOT have a combined row of cells for total defects.

0April 3, 2012 at 12:48 pm #192943

MBBinWIParticipant@MBBinWI**Include @MBBinWI in your post and this person will**

be notified via email.@jashouri – Do you have just the 2 categories (auto and sales)? How many samples in each and how many defects in each? And you only count defectives, not defects (any return for any reason is a defective)?

You’re probably good with a Chi-sq if the number of levels is low (2 categories with whole number sample size and whole number defects would likely fit the bill).

0April 3, 2012 at 11:34 pm #192945

JAshouriGuest@jashouri**Include @jashouri in your post and this person will**

be notified via email.Thank You to both of you. Below are my results and I will describe what I did

All I did was take the Defective count for Sales and the clean ones vs the Defective count for AUTO and the clean one and ran the Chi Square. I did not have more than 5 samples etc…I just took TOTALS. Is that wrong? The P Value is below .05 so this confirms there is a statistical difference. I do have the totals split up by month if that means anything or equates to more than 5 samples…Did I do this correctly and is my assumption correct? See below the chi square results

Clean_1 Defective_1 Total

Sales 2043875 19581 2063456

2041800.24 21655.76

2.108 198.776Auto 1312743 16020 1328763

1314817.76 13945.24

3.274 308.682Total 3356618 35601 3392219

Chi-Sq = 512.841, DF = 1, P-Value = 0.000

0April 4, 2012 at 5:14 am #192951

MBBinWIParticipant@MBBinWI**Include @MBBinWI in your post and this person will**

be notified via email.@jashouri – what you did was exactly right. If you want to peel the onion further, you could split these up by relevant “by” variables (you indicate that month might be appropriate).

0April 4, 2012 at 9:12 am #192953

jamilParticipant@jashouri**Include @jashouri in your post and this person will**

be notified via email.Thank You. Last question. So the p-Value shows there is a statistical difference between if Sales send out a sample vs the sample being AUTO shipped. So the conclusion – is that this is a root cause correct? Are there other tests to do statistically? Or do I now peel the onion further and dig deeper into EACH PROCESS? I guess I am asking – now that it is statistically different – what are the next steps? I GREATLY APPRECIATE THE HELP

0April 4, 2012 at 11:25 am #192954

MBBinWIParticipant@MBBinWI**Include @MBBinWI in your post and this person will**

be notified via email.@jashouri – This only shows that there is a statistical difference (at 95% CI) that there is a difference in expected levels vs. actual. You will need to figure out why using different tools (5 whys would be good here). If you have other by variables (such as month) then you may want to evaluate those similarly and see if there is anything that comes out (maybe there was a shift during a certain month – if so, then you can investigate what happened in that time frame). Right now, look at what’s different between auto and sales.

0April 4, 2012 at 11:29 am #192955

MBBinWIParticipant@MBBinWI**Include @MBBinWI in your post and this person will**

be notified via email.@jashouri – you might also want to Pareto the reason for the return. Your sales people have fewer than expected defects, so they are probably culling out non-productive customers, so find out why.

0April 4, 2012 at 2:52 pm #192961I dont really like to read, this might have been said. And you already have two big fish on the line….

But One of my pet peeves is jumping to a stat tool early. Are you working with a percent or did both groups have the same amount of samples. Regardless If you want to see if there is a difference and you know that one group is 500 and the other is 3 or 90% vs. 7%. Stop. Or do some graphical work. Stop. Then, if you still have questions jump in a more advanced tool.

0April 4, 2012 at 6:31 pm #192978

Mike CarnellParticipant@Mike-Carnell**Include @Mike-Carnell in your post and this person will**

be notified via email.@jashouri You still seem to be at a very high level. Do you have data on where, when or how they were sent?

0April 10, 2012 at 1:39 am #193013

Ravindra JoshiGuest@ravindra.joshi**Include @ravindra.joshi in your post and this person will**

be notified via email.@Jashouri – Do you have the accessibility of how , when & under what conditions thses cans are produced. ARE the cans sent by AUTO SHIPPED & Sales manufactured under same conditions, if yes then probably you can do the GRR as well in my opinion to find out is the defect because of variations during producing cans, u might pick up only defect cases & do the study on the same. Meanwhile as suggested by others if u have reasons please go ahead with Pareto, I will also suggest you to do DFMEA ti see where all the design can fail.

0April 11, 2012 at 5:32 pm #193020

Mike FisherGuest@mfisher**Include @mfisher in your post and this person will**

be notified via email.2 proportions test could be thrown into the mix as well…

0 - AuthorPosts

You must be logged in to reply to this topic.