1
$\begingroup$

I'm having a problem with my dissertation.

My research is "level of satisfaction in relation to customer service: comparative study between 2 fast foods

my questionnaire consists of 19 questions. I'm using SERVERF, the 5 dimensions. 8 questions for Tangibility, 3 for Reliability, Responsiveness, Assurance and 2 questions for Empathy. Example in Tangibility, "Personnel's appearance" the respondents will choose between 1 to 5. 5 - very satiesfied to 1 very unsatisfied.

I have dissected my data, results of survey according to age, gender, year level and schools.

Example, in gender, female, I have the mean per question, example Q#1 and so on, average mean of all the dimensions, exmple tangibility (Q#1 to Q8) then other dimensions, WAM of female, overall satisfaction to fastfood 1 & 2. then male... My respondents are students whose customer of both fastfood.

my question is, how to know if there's a significance between the two establishment according to age, gender, year level and school.

I've tried using regression analysis, dependent variable is Gender then independent variable are the overall satisfaction of my respondents in the two fast food.

SPSS gave me the sig. of gender and fastfood 1; fastfood 2 and gender.

my null hypothesis is "There is no significant difference in the overall level of customer satisfaction in relation to customer service between the 2 fastfood"

I want to know if there's a significant between the 2 fastfood when it comes to gender. is that possible? in spss?

btw, I know that my data is ordinal because I used likert scale. also checked if my data is normally distributed, it's not. so non-parametric.

I'm new to spss. been using it for a week now. can someone help me? Thanks

$\endgroup$
  • $\begingroup$ A distribution that's non-normal isn't "nonparametric"; procedures can be. If it's a Likert item it can't be normal, there's no need to check that. However, if you have summed Likert items to form a scale, you already assumed interval scale when you summed the items. $\endgroup$ – Glen_b Sep 27 '14 at 15:19
  • 1
    $\begingroup$ @Sesese also: normally distributed data is more or less irrelevant for multiple linear regression. normally distributed errors are one of the assumptions underlying OLS. $\endgroup$ – Alexis Sep 28 '14 at 1:38
1
$\begingroup$

First, make sure you have clearly defined your problem of interest, and correctly formulated your null hypothesis. Given that you research is "level of satisfaction in relation to customer service: comparative study between 2 fast foods", the null hypothesis would be "there is no different in terms of satisfaction between the 2 fast foods".

Then the independent variable is the level of satisfaction. It should be a ordinal variable with different levels, which-for example-can be labelled as very satisfied, satisfied, neutral, unsatisfied, very unsatisfied. The "two fast food" is THE independent variable of interest. It can be coded as a group factor. Then other measured variables related are nested within this group factor. Last, "Gender", "Age", etc., are common covariates. Think of the ANCOVA.

You can use the Cumulative-logit model. Basically, it's the logistic regression where the dependent variable has multiple ordered categories. Moreover, a more parsimonious model is the Cumulative-logit proportional odds model. The difference between these two is that the latter assumes a constant slope across all cumulative logits. This is called Proportional Odds Assumption, and of course needs to be justified.

I don't use SPSS, thus cannot give any suggestions regarding coding. But if you use SAS, the PROC LOGISTIC does exactly what you want. You can specify the "link=Clogit" to fit a cumulative-logit proportional odds model. And SAS provides some tests for checking the assumption.

$\endgroup$
  • $\begingroup$ thanks! I'm now checking for SAS. but..How about if i'll only gonna compare the 2 fastfoods, no group factors like gender etc. compare the results of the 2, what will you recommend to use? thanks again :) $\endgroup$ – Sesese Sep 28 '14 at 10:49
  • $\begingroup$ I am not sure I get you. But age, gender, etc, are as covariates, not group variables. They are values associated with each observation(row) in your sample, aren't they? $\endgroup$ – Aaron Zeng Sep 28 '14 at 15:33
2
$\begingroup$

First, this

my null hypothesis is to know if there's a significant difference between the 2 fasfoods in terms of gender, age, year level and school.

is not a null hypothesis. The null hypothesis would be something like "There is no difference between preference for fast food based on gender age etc.

Second you say

I have 5 factors, 1st factor consists of 8 variables, 2nd, 3rd and 4th consists of 3 variables, and 2 variables for 5th factor. Total of 19 variables/questions.

What are these factors? How will you use them? You don't mention them again in your post

Next, the dependent variable should be the rating of the food. I hope this is on a nice simple scale like 1 to 5. The independent variables are age, gender etc (possibly the factors go here, but it is hard to say)

The statistical method, if the rating of food is ordinal, would be ordinal logistic regression. At least, that's where to start.

$\endgroup$
  • $\begingroup$ sorry, my null is "There is no significant difference in the overall level of customer satisfaction in relation to customer service between the two fastfood" I'm using SERVPERF which consist of 5 dimensions. using also 5 point likert scal.e 5 very satisfied to 1 very unsatisfied $\endgroup$ – Sesese Sep 27 '14 at 14:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.