Any help would be greatly appreciated

#1
I am desperate for help as my dissertation has to be handed in very soon and i am totally baffled with the data analysis section.

I have carried out a survey using a Likert scale to measure opinions of Employers, Graduates and Students regarding undergraduate quantity surveying degree course content. I listed each course module then asked them to rate the importance of each one.

I had 8 points on my Likert Scale ranging from Extremely Unimportant to Extremely Important and i asked each group to state how important each module of the degree course was for them.

I have tabled the data and i was told by my tutor to use a Chi Squared test to test the null hypothesis.

My literature review basically concluded that each group of respondents has differing views as to what is important to enable graduates to work as quantity surveyors.

The null hypothesis is therefore 'employers, graduates and students have the same views as to the importance of current degree course content'.

My table has three rows one for employer, one for graduate and one for student.

I prepared a bivariate table showing the observed and expected responses and then i prepared a chi square table showing the chi square value for each response then i totalled this to get the total chi squared value for each course module.

I then calculated degrees of freedom at 14 (r 3-1)x(c8-1).

I used a probability error threshold of 0.05.

This gave me a critical value of chi squared of 23.685.

Each module's total chi squared value was higher than this and therefore i rejected the hypothesis for each module and therefore overall it was rejected.

Now for the problem!!

My problem is that i have got 14 employer responses, 9 graduate responses and 32 student responses but some of the reponses on the Likert scale have not been chosen at all i.e There were no employers choosing response 1 for Module A so the observed frequency was nil. There were also no graduates or students choosing this response for this module so the overall observed frequency for 1 on the likert scale was nil and therefore the expected frequency is also nil.

I have read that for Chi Squared the expected frequencies need to be 5 or more and therefore the tests i have done are useless.

It has taken me ages to do these test and i am hoping i can still use them in some way or is there another test i should be using?

Also i have prepared bar charts showing number of respondents both numerically and by percentage. I have also used bar charts to illustrate the responses to each module. I have used scatter graphs to illustrate the distribution of responses also. I have shown the median and the mode as I have read that the mean is not correct for ordinal data.

Does anyone have any idea if the chi squared tests I have done are useless and if so what I should have done and also if I have done enough with regards to presenting my data in charts etc ?

Sorry this is so long winded but I am very confused so thought I had better explain myself fully!!
 
#2
What type of data does your instrument provide? Your opinion survey . . . does it produce a total opinion score by adding the rankings of each question/item? If yes, you should be able to use ANOVA with post hoc to compare the opinions of each group and look for significance between the groups. IF ANOVA is appropriate, then 2 of your n values are small. Ideally, you would like to have about 30 for each group.

I am confused as to why Chi Square, which is for nominal data? You data seems to be interval.

Just my thoughts.


rs
 
#3
Hi,
I am agreeing with strebeck that here chi-square will not be appropriate test,as your sample size is less in two groups you can try with kruskal wallis H test which is non parametric test equal to Parametric test ANOVA but it is less power ful than ANOVa

regards
Re.Vinaitheerthan
 
#4
Thanks for the help. I have not scored the responses, i was told just to count the number of respondents choosing each response and treat the data as ordinal