You are Here: Home >< Maths

# Need hel with statistical significance watch

1. Considering the following table:

Across the top are different discussion topics, and vertically are political parties. The numbers in the cells represent the number of times that party speaks on that topic.

I want to find out if there is a statistically significant difference on what each party focusses on or whether all parties discuss the same topics equally.

I would use a chi-squared test, but I'm doing the same for 4 other tables. A couple of these tables have an expected value less than 5 (see lib dems) hence breaking one assumption, see below:

Which test would be applicable to all of my tables when testing for statistic significance and whether the parties vary by topic?

Many thanks
Attached Images

2. Anyone?
3. (Original post by Jamie S)
Anyone?
Sorry, I managed to miss this post yesterday! The usual advice that you will see is that the Chi-squared test is unreliable if expected cell sizes occur with values less than five. (I believe this is still widely taught at school) and one is often directed to the Fisher exact test in these cases. The actual situation is more complicated than that, and the chi-squared test is often much more resilient than that simple rule suggests.

In real life, we tend to use chi-squared tests where the p-values are worked out by monte-carlo simulation from the set of all tables with the same row and column sums as the table under consideration.

So, first question: is this for a school or university assignment? If so, we may have to be a bit more sensitive to the out of data advice alluded to above!

If this is "real" work, then we can apply a vanilla chi-squared test, a monte-carlo chi-squared test, and a Fisher exact test. If you do that with the table you have given, the p-values all come out at about 0.08-0.09 - in other words the different ways of doing it are all saying the same thing: no evidence to reject the null hypothesis on no association between rows and columns.

BTW, I'm not sure I what you mean by observing that "but I'm doing the same for 4 other tables". Are you asking about whether a correction for multiple testing should be done?
4. (Original post by Gregorius)
Sorry, I managed to miss this post yesterday! The usual advice that you will see is that the Chi-squared test is unreliable if expected cell sizes occur with values less than five. (I believe this is still widely taught at school) and one is often directed to the Fisher exact test in these cases. The actual situation is more complicated than that, and the chi-squared test is often much more resilient than that simple rule suggests.

In real life, we tend to use chi-squared tests where the p-values are worked out by monte-carlo simulation from the set of all tables with the same row and column sums as the table under consideration.

So, first question: is this for a school or university assignment? If so, we may have to be a bit more sensitive to the out of data advice alluded to above!

If this is "real" work, then we can apply a vanilla chi-squared test, a monte-carlo chi-squared test, and a Fisher exact test. If you do that with the table you have given, the p-values all come out at about 0.08-0.09 - in other words the different ways of doing it are all saying the same thing: no evidence to reject the null hypothesis on no association between rows and columns.

BTW, I'm not sure I what you mean by observing that "but I'm doing the same for 4 other tables". Are you asking about whether a correction for multiple testing should be done?
I'm a research assistant for a Professor of Political Science at the LSE. This result (depending on how interesting the findings are) will be used firstly as an appendix to an Academic Book, and secondly may be published independently in a number of Political Science Journals.

I am a political scientist by training, not a statistician so would not know very well how to do more complex tests - hence would be thankful if you could advise, teach or explain those to me. Although I do have access to 1) Stata 2) SPSS and 3) the internet tutorials. If worse comes to worse I can always contact the Stats department.

As for more details on the study, I would prefer to message you directly.
5. (Original post by Jamie S)
I'm a research assistant for a Professor of Political Science at the LSE. This result (depending on how interesting the findings are) will be used firstly as an appendix to an Academic Book, and secondly may be published independently in a number of Political Science Journals.

I am a political scientist by training, not a statistician so would not know very well how to do more complex tests - hence would be thankful if you could advise, teach or explain those to me. Although I do have access to 1) Stata 2) SPSS and 3) the internet tutorials. If worse comes to worse I can always contact the Stats department.

As for more details on the study, I would prefer to message you directly.
I'm happy to give you some advice on here, but it strikes me that if this work is important and with high visibility, it might be a good idea to have a chat with a local statistician. You may not need to go to the Stats department, though - I would have thought that LSE Political Science would have a few people adept at this sort of quantitative analysis hanging around!

### Related university courses

TSR Support Team

We have a brilliant team of more than 60 Support Team members looking after discussions on The Student Room, helping to make it a fun, safe and useful place to hang out.

This forum is supported by:
Updated: June 12, 2016
Today on TSR

### Edexcel C3 Maths Unofficial Markscheme

Find out how you've done here

### 1,000

students online now

Exam discussions

Poll
Useful resources

### Maths Forum posting guidelines

Not sure where to post? Read the updated guidelines here

### How to use LaTex

Writing equations the easy way

### Study habits of A* students

Top tips from students who have already aced their exams