To properly implement the Shapiro-Wilk test you would need code equivalent to Algorithm R94 [Appl. Statist. (1995) Vol. 44, 4]. If you are comfortable with compilers, you can either compile the Fortran code or use f2c and then compile the C code. Converting this into IGOR code is possible but not trivial.
For v.8, does functionality exist for alpha values other than the default, which I assume is 0.05? If not (yet) is there a dirty way in which to effect this?
I think there is a misunderstanding here that may be due to the examples in the documentation. The test as implemented returns two values: V_statistic and V_prob. At this point there is no assumption about alpha.
After you execute the operation you compare V_prob to the desired alpha value and if V_prob<alpha you reject the null hypothesis that the population is normally distributed. I see no point in adding an alpha input to perform the test V_prob<alpha.
The examples in the documentation are somewhat extreme in that p=0.846139 is simply likely to be larger than any practical value one might use for alpha and p=1.7979e-05 is likely smaller than typical alpha values.
June 17, 2013 at 10:56 am - Permalink
But is there any chance, that this widely-used Test may be implemented in any future Igor version?
July 9, 2015 at 05:33 am - Permalink
More than a chance, it is in IP7 as StatsShapiroWilkTest.
A.G.
WaveMetrics, Inc.
July 9, 2015 at 11:01 am - Permalink
For v.8, does functionality exist for alpha values other than the default, which I assume is 0.05? If not (yet) is there a dirty way in which to effect this?
October 14, 2019 at 04:17 pm - Permalink
I think there is a misunderstanding here that may be due to the examples in the documentation. The test as implemented returns two values: V_statistic and V_prob. At this point there is no assumption about alpha.
After you execute the operation you compare V_prob to the desired alpha value and if V_prob<alpha you reject the null hypothesis that the population is normally distributed. I see no point in adding an alpha input to perform the test V_prob<alpha.
The examples in the documentation are somewhat extreme in that p=0.846139 is simply likely to be larger than any practical value one might use for alpha and p=1.7979e-05 is likely smaller than typical alpha values.
A.G.
October 15, 2019 at 10:50 am - Permalink