Re: Quantity of Samples (Var and Attri) vs C/I and T/I
Hello guys,
... the rule of thumb in which one should have a minimum of 35 samples for variable testing and 60 samples for attribute testing.
well, rules of thumb are difficult since they are more urban
myth than
rules. Sample sizes are dependent on the statistic you are estimating, the accuracy of the estimate, the alpha risk (otherwise known as the "confidence which is 1-α), and the standard deviation of the individual values in the population. This urban myth arose from the days of no statistical software, excel or calculators...
Is there any magic formula that explains this sampling criteria?
No more than as described above. Look up sample size formulas - I believe I posted these a few days or weeks ago... However most of these 'rules of thumb' are derived from rough cuts based on the above formulas and the rough amount of variation around the statistic and the non linear incremental increase in accuracy given the cost. Typically rule of thumb sample sizes should only be used for the first look at data - and confidence intervals should always be applied to these (and all) estimates - and subsequent analyses should use the knowledge gained from the first data to calculate future sample sizes.
What is the connection to Confidence Interval?
Rule of thumb sample sizes are only weakly connected to confidence intervals as stated above. However, you can and should apply confidence intervals to all such estimates. once you have the data you can calculate exact confidence intervals.
What is the relevance if 95/95 Tolerance Interval is the firgure of choice?
Seem like I came across this in one of my training but I cannot remember.
It's the same relationship...these are rough cut sample sizes. you can still apply 95/95 (or other) tolerance intervals to see how 'good' your estimate was using these sample sizes.
in summary these are urban myth. in some cases these will yield gross oversampling and in some cases gross and potentially misleading under sampling. applying the confidence ro tolerance intervals after collecting the data protects you from this.
With today's software including excel, there is no reason not to calculate the necessarry sample size before you sample. and dont' forget that the accuracy of samples is dependent on the representiveness of the sample as much as the sample size. Dewey defeated Truman in 1948 not becuase the polling was the wrong sample size, but because the polling was not representative.