Newb Question - Chi-squared - I love numbers but theses stats are tripping me up

[chese79]

I am look at results of a likert scale and trying to establish statistical significance of the difference in percentage of 'top box' scores. Since the frequency distribution of the likert scale is non-normal I figured categorizing the top box is the best test for this.

I am pretty sure I need to use chi-sqared test for goodness of fit but I am unsure of what my expected values should be?


Count of people who are consider the package appealing scoring the top box:

Yes no
Control package 339 659
Test package 119 220

First am I using the right test? Are my expected values my control package? If not what are they?

 

[kabooze]

Since you have a 2*2 table only you could simplify things (and not bother about expected frequencies) and calculate the chi^2 value with the formula:

N*(ad-bc)^2 / (a+b)(a+c)(b+d)(c+d)

Where N is the total sample size ande the lower case letters refers to each filed of the table:
a b
c d

 

[chese79]

In my full dataset I have nine packages. Would that work with a 9*2?

 

[trinker]

I am pretty sure I need to use chi-sqared test for goodness of fit but I am unsure of what my expected values should be?

The expected values are the values of the cells if all cells had equal proportions.

Have you ever done Chi Squared by hand?

Seeing/doing it helped me understand what's going on. You'll understand expected values much better.

I suggest you do it out by hand. Here is a link (click here)

 

[chese79]

thanks trinker, made sense now.

 

[noetsi]

Why are you doing a chi-square by hand The software will tell you what the expected value is.

 

[Dason]

Why are you doing a chi-square by hand The software will tell you what the expected value is.

Because like trinker said - at least knowing how to do it by hand is enlightening. If most people actually knew what they were doing with an analysis and at least had a general idea of why it works I think things would be a lot better.