I'm not quite sure what you mean by multivariate analysis
I mean as in the tests done May 16, September 20, and October 9 reported at http://meta.wikimedia.org/wiki/Fundraising_2012/We_Need_A_Breakthrough without adjusting the best performing pull-down delivery combined banner/landing page from the beginning of this month (although I don't think we will need the one that follows vertical scrolling. It may produce 30% but that will be nothing if the remaining ~300 appeal messages are tested, unless they don't fit the lognormal distribution that they appear to.)
That would be complex, and could be a disaster...
What are the possible failure modes?
On Dec 28, 2012 9:46 PM, "James Salsman" <jsalsman at gmail.com> wrote:
How about for the April fundraiser, instead of setting a dollar value goal, we agree to use multivariate analysis instead of A/B testing to optimize the messaging from volunteer submissions in advance, then run the whole thing for a fixed time frame, say three weeks, and then use the actual amount raised to decide whether salaries should be competitive with area tech firms, whether Fellowships should be jettisoned, how much personnel to put into the Education Program and engineering, and how much of a reserve to invest, preferably with low risk instruments which pay above the rate of inflation?
I'm on a train so this will be brief... But let's say we do what you suggest, and it only raises 90% of what is needed. What then?
I've had to pare back WMUKs budget in the past few weeks to what the FDC granted. It's difficult enough to do on a smaller budget, I dread to think how the WMF would do it if they had to. A large organisation lucky enough to have a good income should use that income to plan ahead, rather than using fundraising as an artificial cap. You do raise good points on some subjects though - eg workers remuneration, which is generally poor in the US - but it's very difficult to discuss that via a public mailing list. On Dec 28, 2012 10:12 PM, "James Salsman" jsalsman@gmail.com wrote:
I'm not quite sure what you mean by multivariate analysis
I mean as in the tests done May 16, September 20, and October 9 reported at http://meta.wikimedia.org/wiki/Fundraising_2012/We_Need_A_Breakthrough without adjusting the best performing pull-down delivery combined banner/landing page from the beginning of this month (although I don't think we will need the one that follows vertical scrolling. It may produce 30% but that will be nothing if the remaining ~300 appeal messages are tested, unless they don't fit the lognormal distribution that they appear to.)
That would be complex, and could be a disaster...
What are the possible failure modes?
On Dec 28, 2012 9:46 PM, "James Salsman" <jsalsman at gmail.com> wrote:
How about for the April fundraiser, instead of setting a dollar value goal, we agree to use multivariate analysis instead of A/B testing to optimize the messaging from volunteer submissions in advance, then run the whole thing for a fixed time frame, say three weeks, and then use the actual amount raised to decide whether salaries should be competitive with area tech firms, whether Fellowships should be jettisoned, how much personnel to put into the Education Program and engineering, and how much of a reserve to invest, preferably with low risk instruments which pay above the rate of inflation?
Wikimedia-l mailing list Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l
On Dec 28, 2012 10:12 PM, "James Salsman" jsalsman@gmail.com wrote:
I'm not quite sure what you mean by multivariate analysis
I mean as in the tests done May 16, September 20, and October 9 reported at
http://meta.wikimedia.org/wiki/Fundraising_2012/We_Need_A_Breakthrough
without adjusting the best performing pull-down delivery combined banner/landing page from the beginning of this month (although I don't think we will need the one that follows vertical scrolling. It may produce 30% but that will be nothing if the remaining ~300 appeal messages are tested, unless they don't fit the lognormal distribution that they appear to.)
But what variables do you want to test? You've only talked about messages.
wikimedia-l@lists.wikimedia.org