Meh. We chose this place as the venue. That's selection bias. You sure do love semantics.

Mostly it's diditclear, I'm just an occasional instigator and contributor of ideas.

Lolz. What can you do.

Not to be too pointed, but well established rules of statistics and professional norms are not "semantics" nor are they "Meh" Your statement above shows a very fundamental lack of understanding of statistical sample selection. Choosing to post links on FT is not making a sample selection in any way shape of from.

I posted a link, if you bothered to read it I don't think you would be saying "semantics." Its a problem that can't be overcome. You can ignore it, pretend it does not exist, but it is still there.

The data is interesting, but then so is "poll" of Fox watchers on whether Obama is a communist, or global warming is a hoax. None have any statistical validity as to what is happening in the country, or with UA upgrade rates.

This is demonstrably false.

Demonstration:

You said that no one except people who are flying on Y/B/M's are getting upgrade rates in the 80-90% range.

There are people on DidItClear who have. (I think; I haven't actually checked, because I'm lazy.)

EDIT to add a less snarky response: I've dealt with the misimpression that self-selection kills the cat before, again in the context of brain imaging, psychology, etc. studies. Doing statistics was my job.

There is a group of people that insist that those studies can never have any validity because many of them require the subjects to affirmatively seek them out. The fact is that there is a space between that and Fox News. I think DidItClear is in that space.

When you do a statistical test, you calculate a p value that is supposed to represent the likelihood of the outcome you observed occurring in the presence of the null hypothesis. There's always some possibility that random chance produced the effect -- you're just looking to make sure that that possibility is very small. DidItClear can absolutely be used to make such assessments.

Here's an example of how that would work, taken to the extreme. Let's say that I have a hypothesis that at least 10% of 1K's are obtaining CPU's on at least 80% of their flights. Let's also say that I know that there are 1,000 1K's out there. Let's say I pull up DidItClear and it shows that 50 1K's have reported their stats, and they all have CPU rates of at least 80%. That effect could come from one of three places: (1) facts about the entire population, (2) self-selection bias, or (3) random chance. There is sufficient information in this example for a statistical model to assign the likely amounts at which each of those factors contribute to the effect.

To your snark response, its not what I said.

To the rest, I don't know why you keeping rattling on about brain imaging, but no one with even a passing familiarity with statistics would be saying what you are saying.

Throwing out the term P value does not give you street cred; a P value is used to determine the level of certainly that the no interaction or null hypothesis (loosely called as chance) is not correct. It assumes that the data was randomly selected. Absent this you simply can't do a valid P-value, or any statistical test. The idea you could is so out there, that its hard to find discussion of it. Closest I could find in a quick search is the responses to someone who asked if you could. See http://www.talkstats.com/showthread....on-random-data

Perhaps read some more, or consult someone who you know who has a statistical background. They will hopefully quickly answer your questions.

This is what you said:

That's not what a P value tells you.

In any case, my point was simply that many experiments have bias (including self-selection bias), and all we're ever trying to get to is some measure of the likelihood that the observed pattern would happen from chance as opposed to from the null. No experiment conforms perfectly to the underlying assumptions used in statistical analysis. Doesn't keep us from drawing conclusions.

Point to any published peer reviewed study or article anywhere that has drawn statistical conclusions as to the likelihood the association claimed was real vs being random (either a t-test or a p-value) where the data was collected by people self-selecting themselves into the study, or better yet simply posting on the internet. You will not find anything that has been peer reviewed that does this in any field, nor will you find any authoritative source that says you can do these tests on such data. Bottom line, any "statistics" you pump out of a self-selected set of data is junk science.

I look forward to seeing what you come up with, I always like to be shown to be wrong.

and p.s. As I have said the information is interesting, I look at it, I've posted data, but don't delude yourself in anyway it is statistically significant or valid, it it not.

FWIW, my CPU rate this year so far is 82% (upgraded 23 out of 28 flights) but most of my travel is weekend travel and many of my CPUs are for Saturdays, but I've been successful using RPUs for transcon flights on Fridays and Mondays.

For the CPUs, roughly 72% of those were on G fares, 18% on N fares, 10% on L fares.

Nearly all of my flights are from BWI and I've been pleasantly surprised with my upgrade rate for BWI-LAX/BWI-SFO.

The stats will be more and more interesting as we get more and more users entering their flights: something I was keen on when we originally designed the upgrade form was to be able to enter more data than simply upgraded/not upgraded.

There is definitively bias on fare code, routes, etc. reported, as these are directly tied to the reporters. But there is no bias on equipment, size of upgrade lists, whether the cabin left with premium seats empty, whether ToD was offered and at which price, etc.

If all fields were always populated, you could draw some interesting conclusions even if only people who never clear were entering their stats.

My longer term goal is not to simply tell you that SFO-LAX has 10% CPU chance for a 1K. Instead, you should tell DidItClear: I'm leaving Sunday 10am from SFO to LAX, and it should be able to pull up all the stats entered for this route and day of week, compare them to the current published timetable, and give you some interesting facts. Given these, you may decide that you might have actually more chance if you SDC to the 2pm flight.

Now, of course, it is tedious to enter all of that information, but I'm trying to improve the experience (ideas welcome!). There are already a few users who take the time to populate most of the optional fields and even enter useful info in the notes section. Big thanks to you! ^

I salute your project and wish it the best, but it's incorrect to state that there's no bias on equipment, size of upgrade lists, etc. This is so because the very FTers who are most likely to contribute to diditclear are, for the most part, also more likely than most UA elites to book flights with the highest expected upgrade probabilities. In other words, however much a "typical" 1K weighs a plane's F/Y ratio when booking flights, diditclear contributors almost certainly weigh the F/Y ratio to a far greater extent. On the other hand, FTers who are happy with their CPU % may well skip over threads that discuss CPU statistics, while FTers who are dissatisfied with their CPU % are more likely to keep tabs on efforts to collect data about CPUs. Given the many confounding, conflicting variables at issue, it's difficult to ascertain with reasonable certainty any meaningful trends about CPU probabilities from diditclear.

I actually think that basic things that many of us do (F/y ratio; checking F seat map, checking F=#) are probably beyond the vast majority of 1Ks, let alone lower level elites. In my firm, of those with 1K or GS status (about 10 of us) I am the only one who books my own travel, everyone else has their AA do it. I can assure you the AAs are not checking the way I am.

IMHE when I have talked to 1ks, their understanding usually does not go past the level of "its hard to upgrade Monday am, Friday afternoon, and Sunday night" or "SFO/EWR suck for upgrades" or "my rate of upgrades is down.

If you are obsessive about, and flexible on when you travel (unlike me, I usually need to travel at a set time and day, only flexibility I have is choosing another airline or routing) I would bet that you can push your upgrade basis by 30 or 40 points or more off the base line. On the other hand, I'm sure that were I not doing what I can my upgrade rate would be 10% lower, in that I would take the flight routing that works best, and not get upgraded as a result.

So I expect the overall upgrade rate on diditclear to be significantly higher than the usual average because anyone compulsive enough to post their upgrades/missed on a website are probably doing all they can to get upgrades.

The posts on this thread that ring true to me are those 1ks who say "my rate is X" and either "I fly a mixture of x/x/x fares and have little flexibility with lots of Monday/Friday flights" or "I fly a mixture of X/X/X fares and have lots of flexibility, I move my travel days to get upgrades and avoid Monday/Friday, lots of my travel is Saturday."

Again, I appreciate the work going into DidItClear, I will continue to post my upgrades there, I find the information very interesting, but I don't think that it can reasonably be claimed to be representative of the "typical" experience it is instead reflective of the experience of the FT hard core who would post/track their upgrades.

Thanks!

I am hoping to grow the user base beyond FTers. The initial seed for the vision on how statistics should be collected had to come from here, since this community combines all of the experts who know best the topic. But there is another community of frequent flyers who are not here, who are still interested in keeping such tabs. The blend of both worlds should un-bias the results, at least to some extent, provided we can reach out to them.

Thanks spin, me too. (I am not affiliated with diditclear. )

I just have to laugh given the back and forth on upsells and UAs strategy. Two data points, all from the last 15 minutes.

(1) booking SFO-PHL-SFO for 5/15-5/17. Got offered "special only now" buy up offer at ticketing. $399 SFO-PHL, $379 PHL-SFO. 20 minutes later (after it actually tickets) I click though seats and faster than you can say "blue light special" the "special, buy it now "only one seat left" offers are $380 SFO-PHL, and $280 PHL-SFO.

(2) booked SFO-SEA for 5/6 (so am in the upgrade window, no R though). Offered upsell at purchase for $279. Then 10 minutes later (this one ticketed fast) it was $166 when I clicked though seat selection.

It just makes me wonder how dumb the folks who invented this scheme all think we are...

I don't know that they think we are dumb...they just think we a too busy and preoccupied to watch our pocket while they try to steal our wallet.

This makes me wonder about the whole "there are four disparate systems involved in upgrades and we have a project to link them all together" discussion from the United DO last fall. Perhaps the buy up offer during booking comes from one place, the offer when interacting with the seat maps comes from someplace else.