Suppose we want to look at which airlines have similar pricing strategies. The data set looks like this:
Variables: Flight Origination, Flight Destination, Airline1 Price, Airline2 Price, ....Airline10 Price.
Data:
Origination: A, Destination: B, Airline1 Price=100, Airline2 = 120, ...., Airline10=95
Origination: A, Destination: C, Airline1 Price=500, Airline2 = 450, ...., Airline10=505
......
I wonder if I can use PROC VARCLUS to accomplish this objective. But as I read, it seems PROC VARCLUS is not used for this purpose.
May be able to simple things down by running two-way ANOVA (if assumptions met; Dependent=cost, then your two categorical variables=independents). Would just create categories for the different flights. Though, you could try to get more complicated if these results are significant, such as controlling for hubs, etc.
Don't forget to add an interaction term and plot data.
More info:
Expected outcome goes like this:
Airline1, Airline5, Airline10 have the same pricing strategies
Airline2, Airline3, Airline6 have the same pricing strategies
....
What I meant by the same pricing strategies is when Airline1 prices high on a origination and destination pair, Airline5 and Airline10 also prices high.
Not sure how the question differs? Do you want to compare all at once, finding differences between airlines in general? Do you just want to compare select flight prices between airlines? Many options differing in difficulty level. If you just cared about a single flight you could just perform one-way ANOVA and then look at the direct comparisions at the bottom of the output.
No I agree with the OP. I don't think ANOVA is the way to go here. Anova is useful for finding differences but that's not what they want to do. I think some sort of hierarchical clustering based on the 1-correlation dissimilarity measure might work pretty well. The tough part is choosing how many groups there actually end up being.
Another option would be to standardize the data and then use something like k-means or k-medoids utilizing something like the gap statistic or silhouette weight to choose what k should be. This would give you a slightly more automated way to choose the number of groups but I kind of like the hierarchical clustering methods a little bit more.
"His programming is malfunctioning. It begins! Get your weapons, he's going to become a killbot!!!" - bryangoodrich
ANOVA will not give me what I want. Having the same means doesn't they price similarly.
Another point is I want to group variables, not observation.
Can I create a correlation matrix of Airline1 to Airline10 and figure out the groupings?
Yes, I was also thinking about a hierarchical model based on their interest in varclus. In my last post, I was hoping they would better describe their question. Could they cluster on each unique trip combination, but could other levels be added based on only on originations and/or destinations, since these could help explain costs. Was never sure how big the sample was and when there are too many levels. If there are too many levels will the model have issues converging?
If you are wanting to group some of the airlines together, you first need to define what you would want to group them based on, then think about the best strategy. You need clear variables or parameters that are reasonable not only to yourself but to others.
I have been very clear on my objective. I want to group airlines based on the similarity of pricing. This is a general question, not planning to go into a full analysis of airline industry. I can make it even more abstract like this:
Item A, Company1 Price=100, Company2 = 120, ...., Company10=95
Item B, Company1 Price=500, Company2 = 450, ...., Company10=505
Posting Permissions
|
|
Reply With Quote