Somebody asked me to run an analysis of their data. They are testing whether rewards card would have impact on sales.
Design of the experiment:
They run an experiment, they invited four stores, 30 respondents given rewards card and 30 respondents given a purchase monitoring card. Two stores designated as the treatment (offers reward to 30 card holders with rewards card for every 10 store visits) and the two designated as the control (offers no reward but will stamp stickers to the card of the 30 respondents for store visit monitoring).
Data collected
They presented me with sales data (baseline and post experiment sales) but this aggregated sales in every stores and not on the purchase history of the respondents. I also asked the store visits of the 60 respondents but they only recorded the visits during the experiment period. So there is no baseline store visits data. There are demographic data as well, age and gender only.
Initial treatment
I run a two factor ANOVA without replication for the sales data they gave me and the results shows that only the treatment and control group is significant but the pre-post test sales is not. My analysis is that the huge variance of the average sales data between treatment and control stores contributes to the variation, but the sales between pre-post experiments, although post sales is higher but not so impressive.
I wanted to use the store visit data of the 60 respondents, the profile and the behavioral questions on loyalty, accessibility and price. So my question is, what alternative analysis/tool can I use to make use of this data?
There are occasions which the best is not to respond in, but, in this case it is evident that the Sr Langue inot support well n a clear way his question does and, in my opinion, he do not know than the qualitative and the quantitative are very different subjects.
Since the pre-post comparison is not significant, the goal in analyzing the store visits data should be to understand why the control and treatment stores are different. The key question there is whether they were different in the first place.
Did you test the interaction between treatment group and pre-post? It seems like your basic hypothesis is about an interaction effect: that there will be a pre-post change in the treatment group but not in the control group.
Maybe a multiple baseline design might be feasible... given the smaller sample size than usual.
Granger's Causality Method exists and, more advanced, exists the method of simple cointegration and multiple cointegracipon developed by Soren Johansen and Katerina Juselius..
Now, the called Method of Multiple Baselin Designs is not mathematics, Do not interest me
Granger, C. W. J. ( 1969 ): Investigating Causal Relations by Econometric Models and Cross Spectral Methods. Econometric, 37, 424-438.
Hi, I would suggest you to do a paired-t test and see if both pre- and post differ significantly. We use paired t-test to check whether the imparted training/education had any impact on for e.g. student knowledge (single measurement variable). The analysis is simple to use on single set of observation but you may have to use two-way ANOVA for a replicated design and if observations involve multiple measurement variables. Paired-t test can help you to simplify for each measurement if you are not interested to know any interactions among factors. Just a thought.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis