If your Likert data are individual items, then , no, you shouldn't use t-test.

If you are combining many Likert items into a true scale, then the answer may depend on your philosophy and how the data came out.  If there are many different answers, there aren't too many observations at an endpoint, and the data distribution in each group is normal-ish, some people will use t-test.

In general it is safest to treat Likert data as ordinal data.  To compare two samples of ordinal data you might use Cochran-Armitage test, ordinal regression, permutation test for ordinal data.   I also think using traditional nonparametric tests (Mann-Whitney) is probably okay, but some people don't agree.

My Likert data are individual items. I mean each question is based on Likert scale, say 100 respondents for each question. Can you explain how can I combine many Likert items into a true scale? Why I am asking this question is because I have seen some researchers applying t-test on the means of likert data ( though I think since likert is an ordinal scale, therefore mean of all responses is meaningless). For example ,dividing data into two sets by Gender and then applying t-test on means of likert scale data.

Dear Harshita

T- test for two independent samples used for scale and normally distributed data, and for ordinal data or non-normal distributed data you can use nonparametric test.

Good Luck

It is common to use t-test for Likert item data, but it's not appropriate.

As you mention, it probably doesn't even make much sense to compare means for Likert items.

A scale is created by summing or averaging several related items.  For example several questions about religious practices and beliefs could be averaged to a "religiosity" scale.

But if you are interested in the individual Likert questions, just use one the tests I listed above.

Yes sure but first you need to combine them into one group. Please refer to this easy tutorial.