24 ms·
That's fair. I have extensive A/B test results with tens of thousands of data points. I'm the only subject, though.
by leephillips 13y ago
That's fair. I have extensive A/B test results with tens of thousands of data points. I'm the only subject, though.
- eli 13y agoMost HTML is sent multipart with a text version too. You might consider setting your mail client to favor that version.
- gwern 13y agoNo, you don't have that. You have tens of thousands of uncontrolled estimates, without any comparison points, or systematic or objective data collection.
- leephillips 13y agoSort of implied by me being the only test subject, no? Let me try again without the confusing humor element: HTML email gives a bad impression to me and people who share my prejudices and attitudes.
- gwern 13y ago> Sort of implied by me being the only test subject, no? Not at all. For example, you could be sent the same email with a difference separated by a period of months, where the 2 versions of the email are directly compared (sort of a within-subjects design). Or the difference could be randomly applied to all incoming emails, and then the overall populations compared (between-subjects). Just because _n_=1 doesn't mean you can't do experiments on yourself - I do self-experiments all the time: http://www.gwern.net/tags/experiments http://www.gwern.net/tags/experiments > HTML email gives a bad impression to me and people who share my prejudices and attitudes. You don't track your clicks. For all you know, you still click more and get more out of the HTML emails, despite your negative attitude.
- leephillips 13y ago"I do self-experiments all the time" I just saw your work on LSD microdosing. I really admire your dedication to finding something out through careful experimental design. This is an example of an experiment where self-blinding was possible, but I don't see how it would be possible in testing my own reactions to something that, by definition, I would need to be aware of. So, yes, A/B testing would not be possible here on myself. That's what I had in mind. "For all you know, you still click more and get more out of the HTML emails" I never click on them, so there is no question here. Really, I don't need a formal testing regime to know that HTML email sends me a bad signal any more than I do to know how I feel about people wearing obscenity-bearing tshirts.