Let say i am going to change my factor, then the outcome will be change as well. The current function is a static function, which is not flexible with the changing of source data.

Thank you.

]]>

Will StatistiXl run under StarOffice?

Thanks ]]>

You provide the component loadings in PCA, which measure the correlation between the components and the original variables.

Why not the equivalent in discriminant analysis (groupings)?]]>

A feature I would like in future editions of StatistiXl is power analysis. In these days of tight budgets and carefully scrutinised research proposals one has to show that sufficient data will be collected to determine an answer to your research question. I am aware that this can be a complicated topic, and that different routines are needed to calculate power for different procedures! Perhaps power analysis is only of interest to bio-medical and psychological researchers?

Lance ]]>

Some small additions would make the multiple regression routines more useful.

1. Tolerance and VIF and perhaps the Durbin-Watson statistic for colinearity of predictors

2. Some support for hierarchical regression, howing and testing the significance of additional variables as they are added to a model. These could be F-tests or the Sobel test.

Thanks again for a great program

Lance ]]>

Thanks!

Erik Kulstad, M.D. ]]>

Many psychologists - perhaps other disciplines that deal with measurement? - use correlation to study the reliability of their measures. It is fairly easy to carry out item analysis in Excel, but not so easy to calculate Cronbach's alpha - the Rolls Royce measure of reliability. Since alpha is a kind of correlation coefficient it would be a worthwhile addition to the multiple correlation section of the correlation procedures - at least for me!

Thanks for this forum!

Lance ]]>

I have some wishes for the contingency table program. Best (but a lot of work I guess) would be a full log-linear models analysis program. However even without this there are some things that could be added that would be useful.

1. With two-by-two tables it might be useful to include the calculation of odds ratios – these are widely used in medical research.

2. McNemar’s test for correlated proportions in 2 x 2 tables could also be included (Everitt, 1977, pp. 20-22).

3. With r x c tables it might be useful to include the calculation of adjusted standardized residuals. These are suggested by Haberman (1973) (see also Everitt, 1977) and are very useful since they can be compared with the standard normal deviate. I often use them as “post hoc” tests to locate the source of association in a table rather than the approach that you call “the divided chi-square”.

4. With ordered r x c tables (i.e., tables where the categories have an intrinsic order) it can be useful to include a test for linear trend (see Everitt, 1977, pp. 51-56).

5. Some measures of the strength of the association in a table may be useful (many journals seem to want Cramer’s V though I find it hard to interpret). Here’s odds-ratios, and Goodman and Kruskal’s lambdas or Kendall’s tau (much used in “ordinal statistics”) would be nice.

References

Everitt, B S (1977). The analysis of contingency tables. London: Chapman and Hall.

Haberman, S J (1973). The analysis of residuals in cross-classified tables. Biometrics, 29, 205-220.

]]>

Cheers

]]>

In experimental psychology with normal subjects it is almost universal that we repeat many (usually orthogonal) treatments on the same subjects, so we end up with a factorial within-subjects design. I know of nobody in my general area (vision research) that does anything else. Yet for some reason those who write statistics programs appear uninterested in this type of design -- being tuned more to the social psychology type of group differences. This is a pity because there is almost no software that does this type of analysis (except the high priced ones like SPSS -- and even they do not do it well and are unable to plot the results with the correct error bars). If StatistiXL had that facilty I predict it would be the favored software in experimental laboratories. Given that it does the between-groups factorial analysis it should not be much of a trick to add this feature which would be much appreciated.]]>

a. calculate pooled SDs/variances

b. compare two standard deviations

c. t-test between percents

d. confidence interval around a percent

e. two sample t-test between percents

f. confidence interval around a mean

g. t-test between means

h. probability of a z-value

i. critical z for a given probability

j. probability of an F-ratio

k. probability of a t-value

Stan

]]>

I understand the problem of having analyses recalculate automatically whenever the source data are changed, but I'd love to see a recalculate button in the output for the user to quickly repeat the calculation after a data change.

Jonathan ]]>

I downloaded the trial version recently and I really like the program in general, however some features make the usage awkward for my work.

1. Missing values are not accepted.

Data frequently includes missing cells. These cannot just be deleted from the spreadsheet, nor can they be replaced by zero values whithout manipulating data contents. It would be helpful if StatistiXL could handle missing values like other programs.

2. Selection of data range

It would be very helpful if colums coul be selected which are not neighbouring. In a large dataset I would like to compare multiple colums (e.g. by correlation) at a time which are not neighbouring. Such selection is possible in Excel.

3. Spearman correlation

It would be nice to select more than just two columns. Other packages can do this.

4. Test for normal distribution

It is awkward that frequency data is required. Other packages transform the data accordingly in one step. In StatistiXL I have to go through frequency distribution first. This again is awkward because zero frequency cells are required at the beginning and the end. This is not done automaticly. The whole proceedure could be improoved a lot.

All other features are very good. Usage is very intuitive and easy to handle (better than in many other programs). Exact p values are calculated in nonparametric tests, very nice!

I will definitely watch further development of the program and buy it once it meets my demands.

Sincerely,

Dr. V. Witte, LMU, Munich

]]>

The use of the Chi-Square with or without Yate's correction, either of Cochran's corrections, still produces an estimator of the probability that the differences in passing, promotion, hiring, etc. rate for whites is different from the rate for non-whites (or blacks or Hispanics, etc.).

Years ago, EEO (Equal Employment Opportunity) experts stopped using Chi-Square when the Fisher Exact probability became programable for larger sample sizes.

When the probability has been calculated, the EEO field translates the probability back to a two-tailed standard deviation. Since 1977, the United States Supreme Court has accepted standard deviations as the basis for saying that a practice, procedure or test adversely impacts a protected group.

It would be nice as a wish that StatistiXL produce the exact probility rather than the Chi-Square for the heterogeneity test.

Any employment discrimination case in the United States that deals with a class or any pattern and practice employment discrimination case may need to aggregate statistics to obtain an adequate sample to evaluate for potential unintentional employment discrimination.

Federal contractors in the United States, more than 50,000, could also benefit from an easy to run statistical procedure to aggregate hiring and promotional statistics. The statistics should be using exact probabilities then converted to two-tailed standard deviations.

Dick Biddle ]]>

This way, as the statistiXL menu's options grow, one could always have the last procedures at hand, saving time. ]]>

1) a word ("yes" or "no" for a significant difference or no significant difference based on a certain "Prob." value threshold)

or

2) a symbol (*, ** or *** for significant differences, very significant differences or highly significant differences based on certain ranges for the "Prob." value).

Just for ease's sake it would be good to have an option for producing this extra column with qualitative or semi-quantitative information using statistiXL. ]]>

I would appreciate an option to produce a simple, single spaced orderd list of the cases.

Thanks

Eric ]]>