Hello Raksmey Sann

Do you mean Classification Results table? According to the attached image?

In these cases, pay attention to the IVs that you put in the model.

One possibility is that the frequency of cases at different levels of grouping variable is not balanced.

Hello Sann,

for me a correct classification of 40% is too low to be accepted in publication.

Morevoer, discriminant analysis needss to have calibration and validation sets.

You have to take another approach to characterize your results. May be by using unsupervised approach such as PCA.

Kind regards,

Dominique

Hello Sann,

If your value of ".40" refers to the rate of correct classification of the model, then the question becomes, how does this value compare with that of a null model (e.g., just random assignment of cases to group)?

The answer to that depends on the number of groups (k) and the per-group sample size (n1, n2, ... nk). You could compute a base rate of successful classification in the absence of any model IVs by using the proportional chance criterion: C(prop) = Sum of (prop_i)^2, where "prop_i" is the proportion of total cases that are in group i; the summation is across all groups.

As an example, consider a 3-group data set, having sample sizes of 50, 40, and 30, respectively (total N = 120). The proportions are 50/120 = .417, 40/120 = .333, and 30/120 = .250. So, C(prop) = (.417)^2 + (.333)^2 + (.250)^2 = .17 + .11 + .06 = .34.

In other words, if we randomly assigned cases to groups, in the designated proportions (41.7% assigned to group 1, etc.), we'd expect to be correct 34% of the time.

Now, compare your model success to that level: 40% vs 34%. If these were the actual values, your model's success rate would be 6/34 = about 18% higher than the proportional chance criterion. Is that high enough of an improvement to be considered noteworthy? This is a judgment call. I've seen recommendations that the improvement due to the model should be at least 25%, but the consequences of misclassification obviously would be important to consider in such judgments.

Good luck with your work.

David Morse Dominique Bertrand Abolfazl Ghoodjani Thank you for your sharing , :)