"dating" vs. actions

[Ehren Foss]

Jake (Yahoo Research guy who helped with social actions tuner) sent me
this article which is a funny/illuminating analysis of what to say and
not to say in your first contact with someone on a dating site.

http://blog.okcupid.com/index.php/2009/09/14/online-dating-advice-exactly-what-to-say-in-a-first-message/

I'm working on something similar with the social actions data, and
what I can report so far is that the sweet spot for title length is
about 4-6 words. Hit rates are significantly higher in that range,
and trail off on either side.

Similar to the above, I'm working on lists of individual terms that
associate strongly (in one direction or another) with hit count, and
hope to do that per action source - so you can tell if people
searching by action source have significantly different tastes.

Ehren

[Peter Deitz]

Hi Ehren,

Your efforts to illuminate the Social Actions analytics are invaluable.
Thank you so much. It will be quite an exciting day when we can approach
our partners and say, "look, if you want people to click on the actions
listed on your platform and distributed across the social web, the title
should be between 4 and 6 words, and here are the statistics to prove
it." Keep up the number crunching coming.

As for the dating article, I will be reading it closely for very good
reason. I've botch my last few in person intros. :)

All the best,
Peter

Peter Deitz
Social Actions
Founder / Executive Director
http://www.socialactions.com

USA: 415-425-7482 | Canada: 514-824-3270 | Skype: peterdeitz

http://twitter.com/peterdeitz
http://www.linkedin.com/in/peterdeitz
http://my.socialactions.com/profile/PeterDeitz

[Ehren Foss]

Thanks Peter, I appreciate the praise! However, I was fairly off in
my first estimate. I wrote up the findings, then revised them, then
realized I'd need to revise them again. It's trickier than it looks
at first, but at least I get to learn some statistics. :-)

http://www.preludeinteractive.com/2009/10/what-title-length-is-effective/

[Peter Deitz]

Hi Ehren,

I have read over your post. There's a lot for us to consider here. As
you conclude, it may still be premature to have any concrete suggestions
for the action sources. That said, I would be curious to see the second
chart showing average 'hits' per title length for the last month only,
and then to compare that chart to the same period in different action
types (ie, donate, petition, volunteer, loan). Is that possible?

[Ehren Foss]

Yes, I think that's a good solution. We'd want to make sure that
there were enough hits and actions represented in the timeslice so
that whatever we find is statistically significant. If there's enough
data, the same analysis could be run for different types of actions,
different sources, to see if there are any differences there. I'd of
course caution against providing an action source with general
feedback like "your titles are too long" if it really doesn't apply to
them.

I think it may be a challenge to find things within the social actions
data, be confident about them, *and* have those things be what the
action sources are interested in finding out - and are willing to make
an effort to 'twiddle the knobs' as it were.

Ehren

[Peter Deitz]

Hi Ehren,

I see this exploration of the metrics / log information as an iterative
/ long-term process. As we get more clear on the conclusions that could
be made and on the data that action sources seek, we should find that
the two overlap in one way or another. Thank you for all the work you
are doing on this front.

[Ehren Foss]

I think the long term view is very wise. Also, making sure we stick
to some good best practices whenever publishing something - be clear
about how the numbers are generated, the assumptions that were made,
and what may still be missing. To wit, I think I should write up a
page for the metrics-over-time graphs discussing each metric and how
it is calculated.

I would also encourage anyone on this list or otherwise to mention the
project to any statistics or math geeks you konw. Even if they can
spend a little time sanity-checking it would be a big help!

Also, the popular packages/platforms I'm aware of are "R" (open
source) and SPSS (not open source/free, but also quite powerful).
Would probably make sense to switch from PHP for the analysis
framework at some point, so if anyone is familiar with R or SPSS do
let us know.

Ehren