python使用wald检验来检验预测变量的意义

Question

I need to do logistic regression for some data, I have obtained some user features such as their post types , number of friends , number of posts , number of uploaded photos and etc, and have clustered these users into several clusters, now, I want to do wald test to test which predictors (from these user features) are significant for predicting the cluster these users belong to, using binary logistic regression, for example, for users in cluster 1, if the user belongs to cluster 1 , the cluster_label is 1 , and other users' cluster_label is zero , I need to use wald_test to choose which predictors are significant for predicting the cluster label, for example, for predicting users in cluster 1, by doing wald test, the number of friends and the number of uploaded photos have the highest wald score, therefore, these two features are significant for predicting users cluster type in cluster 1; maybe for users who are in cluster 2, by doing wald test, the number of posts and the number of shared news are significant for predicting these users cluster label

the pandas dataframe of these data are illustrated as follows, for predicting users in user cluster 1 :

NoPosts... Friends ...  postCluster0_ratio... postCluster4_ratio  cluster_label
 24     ...   89    ...       0.35         ...        0.3              1
 ...
 ...
 81     ...  161    ...       0.2          ...        0.15              0
 ...
 ...

when cluster_label is 1, it means that this user belongs to user cluster 1, when cluster_label is 0, it means that this user does not belong to cluster 1,and I'd like to by doing wald test to decide which predictors(from NoPosts, Frineds...postcluster0_ratio...) are significant for predicting users cluater label, but from

http://www.statsmodels.org/dev/generated/statsmodels.regression.linear_model.OLSResults.wald_test.html#statsmodels.regression.linear_model.OLSResults.wald_test there are no examples for wald_test in python, I do not know what is the required input for wald_test, how to fit, in one word, I do not know how to use wald_test for my case, could you please help me how to use wald_test, it is better for providing me the code

Answer 1

For individual tests (not joint hypothesis) you can use the t_test which is a special case of the Wald test http://www.statsmodels.org/dev/generated/statsmodels.regression.linear_model.OLSResults.t_test.html

The pvalues for the tests whether the parameters are statistically significant from zero are in summary() and are precomputed, see pvalues in http://www.statsmodels.org/dev/generated/statsmodels.regression.linear_model.RegressionResults.html

wald_test is the more general version of the f_test for joint hypothesis which has some examples that work in the same way for wald_test http://www.statsmodels.org/dev/generated/statsmodels.regression.linear_model.OLSResults.f_test.html