I hate feeling rejected! It’s happened to me again. Back in August 2009, I experienced that for the first time at RealClimate. A comment which I submitted on statistical methodology used in the Steig Antarctic paper did not pass their “peer review” and abruptly disappeared.
Today, I posted the following comment at Tamino’s Open Mind on the thread “Combining Stations” – one of a series of posts about processing station temperature data:
You don’t seem to get a lot of technical observations from your regular commenters in a thread such as this so I thought I might offer some comments about the method you are using to estimate the “offsets” for each station.
I don’t know if you are aware of it, but the “optimal” method appears to be a disguised two factor weighted analysis of variance. One factor is the stations (whose coefficients are the “offsets”) and the second is time (in this case months) treated as a numeric variable. This is not immediately obvious from the form in which you formulated the sum of squares that you are minimizing, but some fairly simple algebra can re-express that SS in a way which is more obviously consistent with an anova. The weights for a given temperature measurement are proportional to the number of stations which have available data for that month. I would have run some comparisons for your example in R, but your aversion to posting data and code makes that more difficult. However, I wrote a simple R program for calculating the offsets as well as some scripts for the anova and ran them on randomly generated data with identical results for both methods.
The side benefit of this is that the estimated monthly grid values are calculated simultaneously with the offsets rather than as (possibly sub-optimal averages) of the offset adjusted temperature series. Variability estimates can also be computed more easily as well.
Within an hour the comment dissipated in electron limbo. I fail to understand why.
Was it the fact that I mistyped the phrase that time is ”treated as a numeric variable” (I meant to sat a “categorical” variable)? Was it my observation that there were not a lot of technical contributions by the readers (there were 14 comments in the thread none of which contributed to the specific material being developed by the author)? Or maybe, my chiding remark that Tamino does not reveal his code or data (thereby preventing most of his readers from real participation)? I honestly don’t know.
I should be pretty irritated but I take comfort in the fact that I am now in good company 🙂 – with people like Jeff Id and Lucia, both of whom I believe were summarily deleted from that blog at one time or another.
I had expected a real discussion regarding the methodology – that he would be gratified that someone actually expressed enough interest in the material to do further analysis (with some pretty interesting insights into his procedure), but I now suspect that he did not even understand exactly what I was telling him. My guess would be that Tamino could actually have learned something from such a conversation, but it appears that he is more interested in appearing to be a mathematical “expert” to his small coterie of readers preferring to pontificate in one-sided expositions
I’ll also opine that the optimal method can definitely be called superior, both because it has no dependence on the order in which stations are processed, and because it simply minimizes the sum of squared differences between all pairs of station records.
rather than interacting with his audience in a meaningful and productive fashion to get a better grasp of the science behind it.
Open Mind? What a laugh!