Yield comparison
Six Sigma – iSixSigma › Forums › Old Forums › General › Yield comparison
 This topic has 5 replies, 6 voices, and was last updated 19 years, 2 months ago by marklamfu.

AuthorPosts

July 31, 2003 at 5:40 am #32935
Hello,
I would like to compare two sets of yield data. The yield data
are written in percentage (80% yield means that 100% went in
to a production line and 80% came out).
There are more than 100 yield data for each set.
Is it OK to use ttest for this comparison to see which set has
high yield?
Since this yield = (1defect percentage), I was wondering if I
need to use something else, such as poisson comprison.
Thank you,
Kim0July 31, 2003 at 6:10 am #88502Hi Kim,
It’s an interesting question, the yield data are plotted as negative skew on Normal Probability chart since most of your data fall in the high end between the range from 0% to 100%, so it is not nomal distribution data set. Personally I don’t know if it is correct or not by using ttest (2 samples) to assess the difference, but I think you still can use it, my concern is the “mean” here donesn’t mean anything ( what does 85% average from a data set mean?).
I think another method could be used, Box Plot chart, to see the difference. This is more directly and visible.
I would like also to share any ideas on this questions.
Thanks,
Mike0July 31, 2003 at 6:16 am #88503
Sigma SinghMember@SigmaSingh Include @SigmaSingh in your post and this person will
be notified via email.Kim
By design t test is meant to test the equality of means of “continuous” data. Now – yield is inherently derived from attribute data. how many passing or failing data points (not mentioning how well they passed). Hence chi square test is the right comparision in this scenario.
sigma singh0July 31, 2003 at 7:21 am #88507Kim,
Since yield is measured in % (and therefore continuous), I would vote for t test.
However testing variances (SD) is also very important before applying t test. You could use Box plots for this.
Bee0July 31, 2003 at 1:52 pm #88520Strictly speaking, yield is in fact a discrete metric. However, many projects that I see treat it as a continuous metric to allow the use of better analysis tools. Statistically, this may not be the greatest way to proceed, but it’ll get you to improvement, so I say go ahead. However, if the data is not normal, why bother testing the means? I must agree that Chi Squared is probably the way to go. Or a proportion test.
0August 2, 2003 at 5:40 am #88559
marklamfuParticipant@marklamfu Include @marklamfu in your post and this person will
be notified via email.Yield is typical attribute data , ttest is used to comparing 2 sets variable data, if you want to use Ttest or Poisson for attribute data, I believe, it is not accurate, Generally, we can plot unit(daily, weekly) yield as a trend for comparing, or transfer attribute as a variable data and then use ttest or other tool(s) for comparision,
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.