Hi Community,
I hope someone can help me because my statistical knowledge is relatively weak:
I have a descriptive research project. It is about hotel amenities and their mentioning in reviews and the question "how can hotel amenities be classified with respect to their satisfactory effect on hotel guests?"
My proposed model is that we can group amenities in four groups:
- basic amenities ( mentioned when missing/ not working, not mentioned when existent/working) example: air conditioner (in a hot country)
- excitement amenities ( not mentioned when missing / mentioned when existent) example: video console on the room
- performance amenities ( mentioned when missing / mentioned when existent) example : breakfast
- insignificant amenities ( not mentioned when missing / not mentioned when existent) example: room telephone
I attached an image of how I image the "model_mentioned"
The whole model is an adaptation of the "three factor theory of customer satisfaction" with the exception that my data and model is not cardinal. the amenities are either there or missing. so I can't just make a nice regression.
With regards to the satisfaction impact of an amenity I thought to model it this way: [see attached model_satisfaction]
Explanation:
basic amenities will have a weak to strong negative impact on satisfaction if not present.
excitement amenities vice versa
the lines for the performance amenities are "imaginative" . For a performance amenity I will have a distribution in the first quadrant ( satisfaction / present ) and one in the third ( dissatisfaction / not_present). The length of the imaginative line connecting both should indicate the strength of the performance amenity.
Now here is my big question:
which statistical tests ( especially for model_1 ) do I need to perform to check my model ?
Unfortunately my supervisor can not help me there, so that's why I'm asking you the community. I really hope someone can help.
My reduced data looks like this
review_score | hotel_has_amenity_X | review_mentions_amenity_X |
1-5 |True / False | (-1 -> +1)*
* my text analysis program looks whether amenity X ( or synonyms of it) are mentioned in the review. if not it returns "FALSE" if true then it performs a simple sentiment analysis on the sentence where it occured). I want to treat negative sentiment as "FALSE" since there's no difference for the hotel guest in whether the air conditioner is missing or not working.
Though the problem appears to be a bit troublesome. What I make of the situation is that there are four groups that need to be compared while the DV is customer satisfaction. Descriptively, it will be the comparison of means while inferentially it will be resolved by one way ANOVA.
If all else fails, you could use a permutation test. For this:
(i) choose a measure of how big an effect you think you see in your datal
(ii) evaluate the null distribution of your chosen measure (test statistic) by doing a collection of random permutations of your data and evaluating the test statistic on each.
Here the random permutations need to be done in a way that represents equivalent sets of data if there really were no effect.
Thank you all for your answers!
The chi-square test is indeed perfect to check model_1 I guess. "Excitement" and "Basic" amenities should reject the null hypothesis of independence where as the "performance" amenities should not reject it. Right, M. Cooper
?Concerning the one way Anova I'm a bit in doubt Muhammad Mumtaz Khan but perhaps I'm just misunderstanding it:
Normally you have one independent variable which has 2+ groups of categorical data ( Music = [Rock, Jazz, Rap]).
In my situation I have :
the rating score is the dependent variable. Shall I create the different groups in my case like this:
- (Hotel_has_Amenity_X & Review_Mentions_Amenity_X)
- (Hotel_without_Amenity_X & Review_Mentions_Amenity_X)
- (Hotel_without_Amenity_X & Review_not_Mentioning_Amenity_X)
- (Hotel_has_Amenity_X & Review_not_Mentioning_Amenity_X)
I ask because different to the Music example my groups are not ( basic, excitement, performance, indifference) are just indirectly in the data.
And... in the end this would be like declaring all these groups dummies and then performing a simple regression analysis over them right ?
Thank you guys for all your help !
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis