Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Differences in calculations between CogStat and other programs
Sometimes different programs use different solutions (e.g., different hypothesis tests) for the same question, or sometimes they use a different version of the same solution (see some examples).
Here we list some of the differences in the calculations between CogStat and other statistical programs.
Standard deviation of the sample and the population
New in v1.6.
Software packages (e.g., SPSS Descriptives, or spreadsheet software STDEV or STDEV.S function) often suppose that you're interested in the population estimation, and the Bessel's correction should be used (e.g., even if in SPSS the menu Descriptives ambiguously may refer to the sample). For an uncorrected sample standard deviation in spreadsheet software, use the STDEV.P function.
Range for ordinal variables
New in v1.6.
Range is the difference of the smallest and the largest value. Sometimes it is recommended as a dispersion measure for ordinal variables. However, in ordinal variables the difference of two values is not known, and subtraction is meaningless. In Explore variable, CogStat displays only the minimum and maximum values, but not the range.
Wilcoxon signed-rank test
Many times statistical programs report W or Z as the test statistic. On the other hand, CogStat reports T (because the supporting function from scipy calculates T). Independent of those statistics, the p value should be the same.
CogStat uses the Yates's correction for continuity, which is the default version in Python scipy.stats, similar to R. In SPSS you'll find this version in the Continuity Correction line. See also this description.
There could be other differences between some implementations of the same tests, because different versions of the same tests could be used. See some background about Wilcoxon signed-rank test, McNemar test or Levene test.