So t-test were performed on samples Picking 100 random samples of teachers from the control group that matched the teachers in the treatment group gave different results – sometimes the t-test comparing means for each sample showed that the treatment group was statistically significantly different from the control group and sometimes not. It does not state at what level of confidence the significant difference was calculated for these t-tests. Different levels can produce different conclusions. What is significant at one level is not significant at another. The 95% refers to the number of tests needed to conclude statistical significance (seems a particularly high bar) not the confidence interval for each particular t-test. You will note that all this does anyway is show statistical significance. It is not proof of anything.