Textbook author David Howell has a nice online note on Treatment of Ordered Variables in a Contingency Table (see link below).  You might try the approach he describes there--although with 5 categories, your sample size is probably a bit lower than ideal.  HTH.

https://www.uvm.edu/~dhowell/methods7/Supplements/OrdinalChiSq.html

thank you Bruce Weaver

I will go through it..

This isn't a specific answer to your situation, but a great general guide for choosing a statistical analysis:  http://www.ats.ucla.edu/stat/mult_pkg/whatstat/ 

You might look in to:

ordinal regression;

permutation test;

or traditional non-parametric tests like Mann–Whitney U test.

Your dependent variable is an ordinal variable.

It sounds to me like your independent variable is categorical (nominal, or groups), and has only two levels, but I am not sure based on your description.

You need to consider the nature and number of your dependent and independent variables and the hypothesis you wish to test or the model you wish to build.

The table on this page will help you choose a test based on the nature and number of your variables:

http://rcompanion.org/handbook/D_03.html

Permutation tests are not included there, though --- depending on the software implementation --- might be a good solution.

thankyou Salvatore S. Mangiafico, I tried the traditional non-parametric Mann-whitney test..it worked

The Wilcoxon-Mann-Whitney test was designed with a continuous (or at least scaled) outcome variable in mind.  There is a correction for ties, of course, but the problem with a DV that consists of a small number of ordered categories (e.g., a 5-point Likert-type item) as the DV is that there will be a HUGE number of ties.  I don't think the test was designed for that situation, and am not at all convinced it performs as intended.  That is why I recommended the Mantel-Haenszel 'ordinal' Chi-square, and (possibly) why others are recommending ordinal regression.  YMMV. 

That's a good point, Bruce.  I wonder if anyone's done any work on that specific question.

I did some general playing with a couple of toy data sets comparing M–W to ordinal regression* and linear-by-linear test**. In all cases the ordinal regression had a lower p-value than the M–W, but the M–W didn't wig out, even as the number of ties became the large majority.  The p-values were relatively close, whether small, near 0.05, or large.  The linear-by-linear was actually quite close to the M–W.

This was just playing.  No systematic investigation.

I played just a little bit with Paired Rank-sum, Kruskal–Wallis, and Friedman.  Ordinal regression came up with lower p-values, but nothing unreasonable from the traditional nonparametric tests.

--------------

* Ordinal regression by cumulative link model, R, ordinal package

** Linear-by-linear test, mentioned in Agresti, conducted by permutation test, R, coin package.  Analyzes an ordinal x ordinal matrix.