Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

I have 2 sets of brand preference data (in %) to measure impact of online advertising on target group F20-40 years i.e., 1.different time period (pre/post - not same consumers but same target group) & 2. Control (not seen online advertising) vs Exposed (exposed to advertising). Could you suggest what statistical significance test I should use and why?

share|improve this question
3  
When you say you have 2 sets of brand preference data, is that just another way of saying you have two groups: exposed and not-exposed? When you say your target group is F20-40 years, do you mean that you have one target group or that you want to look at brand preference by age? Assuming the simplest situation, you have a 2 by 2 design: exposed/not by pre/post. You could consider doing a 2-way analysis of variance. Have you looked at the distribution of your brand preference data? If so, what does the distribution look like? –  Joel W. Aug 7 '12 at 17:15

3 Answers 3

If you assume that the preference is a normally distributed random variable, you can use a two sample t-test. It will test whether the null hypothesis that data from two groups are independent random samples from normal distributions with equal means and equal but unknown variances, against the alternative that the means are not equal.

You can use ttest2 in MATLAB.

share|improve this answer
    
We have collected brand preference (in %) data among F20-40 years for both pre & post online campaign. We want to test whether the observed difference in preference is statistically significant? Similarly we have other set of data among F20-40yrs...in this case we have brand preference data using CONTROL (not exposed to online advertising) and EXPOSED method (data is collected during campaign period). We want to test whether the observed difference is statistically significant. Could we do t-test with repeated measures for 1 (pre/post) and t-test independent samples for control/exposed? –  srk Aug 7 '12 at 17:36
    
Whether a certain test (t-test in this case) is applicable depends on the validity of the assumptions the test makes on your data. So, I would first answer the questions that Joel W pointed out in his comments to your question. If the distribution of your data look like the normal distribution, yes you can apply the two-sample t-test to see if there is a significant diff between the exposed and control groups. –  emrea Aug 7 '12 at 20:11
    
thanks for your input. Brand preference is coded as 1="prefer" and 0="do not prefer". 1. Pre and post campaign data are collected at two different time periods with different respondents (but from same campaign target audience i.e., Female 20-40yrs). 2. Control and Exposed experiment is for different brand and campaign but preference is coded similarly. Control and Exposed experiment happens at same time period & with different set of respondents (with same target audience, F20-40yrs) i.e., we ask whether respondent has seen particular campaign, if yes, "Exposed" otherwise "Control". –  srk Aug 8 '12 at 1:50
    
To me, two-sample t-test seems applicable in both cases (if you assume that the percentage preference is normally distributed). –  emrea Aug 8 '12 at 5:30

If I fully understand your problem, I would try a difference-in-differences approach. The basic idea is this (though in your case, I do hope the lines slope up):

enter image description here

You can estimate the effect with a simple regression of brand preference on three variables and any other covariates that you may have collected. You want to test that the interaction coefficient is greater than zero.

share|improve this answer

You could treat this as a 2 by 2 design. Variable 1 is exposed/not and variable 2 is pre/post. People are nested in Variable 1 (meaning that each person gives you both pre and post information for either exposed or not exposed conditions). You could analyze this with a 2-way analysis of variance (ANOVA) with nesting. This is a somewhat sophisticated analysis, so you may not have run into it before. A simpler approach would be to compute a pre-post difference score and then do an independent groups t test on the difference scores, comparing exposed versus not exposed groups. This simpler approach should serve your purpose. Of course, you need to think about meeting the various assumptions of the t-test (or ANOVA).

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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