# Thread: Dissertation help with Likert scale

1. ## Dissertation help with Likert scale

Hello,

I'm having some problems with the stats for my dissertation.

My dissertation is about the effect of the physical surroundings of a soccer stadtium on the spectator satisfaction.
I'm expacting that spectators who are more positive towards the stadium are more satisfied then those who are more negative.

I've created a questionnaire to test this.
First I ask some personal questions like **** age, fan-factor,...
Then I have 32 propositions which they should answer on a 5-point Likert scale. Ranging from Strongly disagree to Strongly agree.

Example:
There is always enough parking nearby the stadium
Strongly Disagree - Disagree - No Opinoin - Agree - Strongly Agree
The seats are comfortable.
SD - D - NO - A - SA
It is easy to get drinks.
SD - D - NO - A - SA
The toilets are clean
SD - D - NO - A - SA
...

Then I have 5 questions where they have to put a mark depending towards which end of the spectrum they lean to more.
Example:
The quality of this evening was:
Horrible * * * * * fantastic
The General feeling that I have right now:
Puts me in a bad mood * * * * * Puts me in a happy mood
....

Now,
The idea is that people who are more positive with the first 32 questions are also more positive with the latter 5 ad are therefore more satisfied.

How do I prove this? Which test should I use?
And are my scales Ordinal of Interval? Because I've been redading about Likert being interval...
I've been thinking about Crosstabs and chi-square, but it's not quite clear to me.

2. Originally Posted by Bunecarera
Hello,

I'm having some problems with the stats for my dissertation.

My dissertation is about the effect of the physical surroundings of a soccer Now,
The idea is that people who are more positive with the first 32 questions are also more positive with the latter 5 ad are therefore more satisfied.

How do I prove this? Which test should I use?
And are my scales Ordinal of Interval? Because I've been redading about Likert being interval...
I've been thinking about Crosstabs and chi-square, but it's not quite clear to me.

Why dont you start by summing the scores of both tests per person.
You then have "pairs" of data,

person...test1...test2
1..........54.......15
2..........75.......18
ect

Then try and correlate t1 with t2 using one of the nonparametric techniques for measuring the degree of correlation for nominal or ordinal data. That should tell you exactly what you need.

E.g. Crammer' V correlation coefficient, the Spearman rank correlation coefficient, the Kendall rank correlation coefficient and others..

read up on them and find one that fits your situation (dont just run them all and choose the one that supports your hypothesis).

hope this helps!

3. Ok, you lost me already.

Take sums of what exactly?

4. Originally Posted by Bunecarera

Take sums of what exactly?
Both tests have questions that rank in score from 1 to 5 (unhappy to happy, disagree - agree). With each question answered you have 32 values that you can sum in test 1 and 5 in test 2. The summation of these values is an estimate of how satisfied your test subject was according to each test.

Once you've done this you'll have two values per person that took the tests.
Then you can correlate these values to see if high scores on test 1 systematically occur with high scores on test 2.

A little bit clearer now?

5. Ok, thx. A lot clearer now.

For which non-parametric test to use, I can't see a big difference between Kendall and Spearman in my case. Only the fact that using Spearman with Likert scales, assumes that the (psychologically) "felt distances" between scale points are the same for all betweens of the Likert scale used.
Which is the case with me. Since the possibility's between which the persons could choose were given in squares of the same size.
If this equi-distance is not justified, Kendall should be used.
So, Spearman is ok for me I guess...

Also, is there a way to use crosstabs? Or this this not necessary for me? With tabs I thought I could see how people answered one question and then see how they answered on one of the last 5 questions. And what relationship exist between them. But I'm not sure how to test this then.

And last, I wanted to use Chi Square to see if the answers on each of the questions were significant and not random. Is that ok?

Thx

En alvast bedankt TheEcologist. Als het in het Nederlands moet, zeg het dan maar. Moest het dan duidelijker worden.
(Thx TheEcologist, if I have to speak dutch, let me know if that makes it easier).)

I'm new, sorry. I have a stats question that is kind of urgent and I cannot seem to start a new thread. anyone? thanks.

7. ## Dealing with missing values using Amos 7

Originally Posted by BHD
I'm new, sorry. I have a stats question that is kind of urgent and I cannot seem to start a new thread. anyone? thanks.
Hi
I am currently working on a data set with 500 cases that has data MAR. I have data MAR, across 100 odd items. I am aware of the Select mean and intercept option in amos for imputing data.
Specifically, How does one impute the data for all cases. Is there any website or reference that states how this is carried out? I have read thu the user guide and the book by Byrne 2001, but steps on imputation are not explicity outlined.
Would appreciate some suggestions on the same
Thanks

8. Originally Posted by Bunecarera
Ok, thx. A lot clearer now.

For which non-parametric test to use, I can't see a big difference between
And last, I wanted to use Chi Square to see if the answers on each of the questions were significant and not random. Is that ok?

Thx

En alvast bedankt TheEcologist. Als het in het Nederlands moet, zeg het dan maar. Moest het dan duidelijker worden.
(Thx TheEcologist, if I have to speak dutch, let me know if that makes it easier).)
Nederlands kan maar soms vind ik het moeilijk om de juiste nederlandse vertaling te vinden van al die engelse termen dat ik dagelijks gebruik als ik met statistiek bezig ben. Bovendien als we in het engels verder gaan dan kunnen andere mensen ons ook volgen. Waar kom je eigenlijk vandaan?

(We can speak Dutch however because when I do stats its always in English I find it difficult to translate everything all the time therefore I think English is better. Plus more people can follow our topic then)

I think correlation is the right way to answer your question, with chi-square you can’t really establish the positive relation between two variables quite as easily. Therefore I don’t think cross tabulation is necessary.

"With tabs I thought I could see how people answered one question and then see how they answered on one of the last 5 questions. And what relationship exist between them"

-> You can still do this with correlations!

Just correlate;
e.g.

Score question 1 with summed score test 2
Summed score questions 1 to 5 test 1 with summed scores 1 to 3 test 2
ect ect ect,

You can test if there is a positive or negative relationship between any question on test1 with any question on test 2. Just make sure you can devise an a-priori hypothesis on why this makes sense, e.g.

Hypothesis: Comfortable seats lead to a better enjoyment of the show because people are less distracted.
Test: correlate seat question scores with enjoyment question scores

The higher the correlation coefficient, the stronger the relationship (make sure it's significant though). One word of caution here: check for collinearity (that two variables from test 1 are also related).
For example if people who think the chairs are comfortable, also systematically think the service is good too (just a lame example). It becomes difficult to distinguish which factor: drinks or service contributes to enjoyment.

9. Allright, English it is then.

The higher the correlation coefficient, the stronger the relationship (make sure it's significant though). One word of caution here: check for collinearity (that two variables from test 1 are also related).
For example if people who think the chairs are comfortable, also systematically think the service is good too (just a lame example). It becomes difficult to distinguish which factor: drinks or service contributes to enjoyment.

First, I don't want to use Chi Square to test for a relation or correlation between questions. Just to test if the answers on each of the questions seperatly are "correct" and not random. You understand?

How can I test for the collinearity? I adapted the questionnaire from an other article from my literature so I expect it to be "good".

And also, I do want to know which question is more responsible for the spectators satisfaction. So which question or factor contributes more to the satisfaction.
How can I test this? Even though it is difficult..

10. Originally Posted by Bunecarera
Allright, English it is then.

First, I don't want to use Chi Square to test for a relation or correlation between questions. Just to test if the answers on each of the questions seperatly are "correct" and not random. You understand?

How can I test for the collinearity? I adapted the questionnaire from an other article from my literature so I expect it to be "good".

And also, I do want to know which question is more responsible for the spectators satisfaction. So which question or factor contributes more to the satisfaction.
How can I test this? Even though it is difficult..
You can use a chi-square to test if the answers differ from an expected frequency. If you assume that in a random scenario each answer is equally likely you can calculate expected frequencies for a certain sample size. If it is necessary to test for this than this is indeed a way in which you can do this.

Which question or factor contributes more to the satisfaction?

You can test which factor explains more of the variation in satisfaction, using correlation. Split your testscore results into your different factors and just look at the strength and significance values of the correlation, that is already a good way to differentiate without using more complex forms of analysis.

Some other options that can give you similar info are ofcourse multiple-regression (also check GLM's) and Multi variate statistics (MVA).

Especially MVA can be a good first step in exploring your data, when you have many factors and a complex dataset.

You can check for collinearity by correlating your explanatory varaibles (factors) with each other (if they have high & significant correlation coefficients they are colinear). That would give you a good idea of what factors reveal the same information. If you are using SPSS for instance, you can get colinearity statistics (e.g. the Variance Inflation Factor ‘VIF’) when running a multiple regression. That the factors in the survey show collinearity could be a deliberate design element and does not immediately mean that it is "wrong". In past analyses knowing which factors are collinear helped me to beter understand complex systems.

The only thing to do is read up on these options. Correlations would be my first step then I would try other options.

Cheers,

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts