These resources can help you:

http://pages.uoregon.edu/aarong/teaching/G4074_Outline/node6.html

Article Possible Errors in Quantitative Data Analyses

In your case, you need to use Multinomial Logistic regression, since you are trying to predict "membership" of more than two categories (in contrast, predicting membership of only two categories would call for binary logistic regression).  The analysis boils down to a set of two comparisons in your case. For instance, since you have three categorical outcome variables (say var1, var2 and var3), you can compare results against the first categorical variable (compare var2 with var1 and var3 with var1, for example), or compare results with the last categorical variable (var2 with var3 and var1 with var3, for example), or a custom category (var2 in this case, for example, compare var1 with var2, and var3 with var2).  The major part of results and output roughly would look similar to that of binary logistic regression, but interpretation would obviously depend on the type of analysis. You can also check the attached link on how to run this procedure in SPSS. Hope it helps.

https://statistics.laerd.com/spss-tutorials/multinomial-logistic-regression-using-spss-statistics.php#procedure

You can use SPSS and  apply linear regression analysis for causal relationship.

I agree with the other comments regarding the data analysis, but just want to add that as soon as you say "causal" relationship, you must first ensure that your data come from a controlled, scientific experiment following the standard rules for designing experiments. Data that come from surveys, and other such methods are correlational or observational and cannot be used to determine causality. Only by experiment can causality be assessed. 

Yes, causaility can be determined by controlled experiment. However, if question means the "degree" or extent of already (theoretically) established causal relationship and you are just measuring the already validated causal effect in a particular context, then survey data can still be used to measure the extent. Of course, in the process, you need to check for assumption violatations of method(s) that you are using.