3 ms·
Ask HN: How many people needed for effective A/B testing?
Hi HN,
I run a review website (similar to Yelp) for a niche business market. Currently, business owners can manage their listings for free, but I would like to make some money now by offering additional features. I don't know if I should offer one premium tier for say $29 a month with all the features, or multiple tiers ranging from $19 to $39 a month depending on what features they want (think: premium, platinum, ultimate).
Before I create any of these features, I'm going to see if my current customers are even interested in paying for them. So I was thinking of doing A/B testing, where half of my customers would be informed about a one tier structure, and the other half about the multi tier structure. But here's my problem: I only have 170 people who are currently signed up to manage their business listing. Is it worth doing A/B testing with such a small population? Or should I just pick one option and go with it, hoping for the best?
TL;DR: Is having 170 people too small of a population to do effective A/B testing?
Thanks!
EDIT: changed the wording to use correct statistics terminology.
- drnex 15y agoi would say, it depends on the results, if from the 170 you get a contundent answer then I would say its a go, if results are not contundent enough you can keep your a/b running until it does
- drnex 15y agoyou should ask a statistian tho
- fleitz 15y agoWell if all you have is 170 people then you're not sampling but testing the population. I'd go with the A/B testing option just to get more experience with it. Having data at a low confidence interval is better than having no data at all. Maybe it's a wash maybe it's actually very important. Worst case is there is no overwhelming winner and you're back where you started. Use the following link to determine the sample size you need. http://www.surveysystem.com/sscalc.htm http://www.surveysystem.com/sscalc.htm
- gs7 15y agoThanks for the link and the correct terminology (I updated my post).
- fleitz 15y agoTo expand it really depends on what the results of the testing are: If you split your 170 customers into two groups of 85, if 40 of 85 pick A and 5 of 85 pick B then it's extremely likely that A is the better choice. If 39 pick A and 40 pick B then you'd need a larger sample size to overcome the margin of error. Of course at a 95% confidence interval any result even in excess of the MOE is pure chance 1 of out 20 times. What I'd do is split your customer base into 3 groups of 50% 25% and 25%, test each 25% and then apply the winner to the 50%.
- gs7 15y agoThat's a great suggestion, thanks a lot!