2022 Marie Curie Postdoctoral Fellowship (HE-MSCA-PF-2022)

[Axe]

I think the start of rating the ENV panel was yesterday. Is there anyone from the ENV panel whose phase change happened in the previous days?
Also, is there anyone who, for example, changed the phase today?

Answers to these questions are very helpful.

I think it started on Wednesday 1st February. From the spreadsheet, couple of proposals were ranked on the 1st


https://docs.google.com/spreadsheets/d/ ... edit#gid=0

[goshgoshgosh]

There are a lot of variables that may determine the timing of the ranking status. Most likely, the system "ranks" the scores in batches. I am not sure why this happens, because we are not talking about big numbers here. But there are many possible explanations of how these "batches" are put together, and possibly they are not completely random (for instance, probably EF and GF are ranked at different times).

One thing we know for sure is that the system does not start "ranking" the scores from absolute highest to the lowest. We know this for sure by reading the sheets the "previous generation" has put together: some people turned into "ranking" later, and yet scored higher than those who turned earlier. To be plain, the system is not told "take all scores, divide them per panel, for each panel, rank them from the highest to the lowest".

What is likely, is that the system is told to select a number of proposals scores on some other basis—very likely, per submission numbers, which appear to be consecutive—and rank them per panel. Something like: "take all the scores from the first one hundred proposals, divide them per panel, rank them per panel; take other 100, repeat..."

I believe that the idea that the first proposals to turn into ranking are more likely to be founded is possibly wrong, because is not the way computer programs are usually designed. The program does not know what scores are to be considered as "higher" until you give it a criterion to rank them in the first place. To obtain such an input, one should design a program such that (for instance) "If scores are => 90%, then rank them". But it seems a very inefficient way to design a program.

[goshgoshgosh]

As it has been said before, there is likely no logic in the timing of the ranking. It seems more or less clear that the ranking process is not done by a machine but by a person (or several of them) per panel in batches for one week.

If it is done by a machine, there is most likely a logic. It is possible, but in this case not efficient to design a program that says "take 100 scores randomly, divide them per panel, rank them according to x". Most likely they are ranked consecutively or according to a certain criterion. Consecutive number, date of submission, or most likely date of approval of the final score by the panel... are all good guesses.

As I write above, the program does not know what score is to be considered as higher until one tells it what the criterion to adopt. So one cannot design a program that says "rank all high scores first, and then proceed with the others": the program must be told what the ranking criterion is, then it ranks all scores together, and only at that point it knows what the highest ones are.

One can tell it do rank first all scores from a certain threshold up (e.g. from 90% up), but although this may be efficient (for in certain panels it never occurs that proposals below 90% get funded, so they can possibly organize their meetings more efficiently, though I doubt they are so well organized overall) we do know this is not the case, because we know from the past that people turned to ranking later although having higher scores than those who turned earlier.

That said, we are talking small numbers here. There is no, strictly speaking, "computational" reason why results are not ranked all at the same time.

So most likely the reason is human.

[lyb]

To me, it is intuitive from last year's spreadsheet that none of the applicants with evaluation to ranking changed on 2nd March or afternoon of 1st March was granted. To be honest, the correlation between the evaluation/ranking changing time and the winning chance is also clear in general. The two things that may challenge these conclusions were the insufficient samples and the uncertainty of the time when they checked the status.

[michelef]

To me, it is intuitive from last year's spreadsheet that none of the applicants with evaluation to ranking changed on 2nd March or afternoon of 1st March was granted. To be honest, the correlation between the evaluation/ranking changing time and the winning chance is also clear in general. The two things that may challenge these conclusions were the insufficient samples and the uncertainty of the time when they checked the status.

What about the folks that remained in evaluation until the end and still got granted? There's at least one in the spreadsheet, perhaps more that didn't add the information

[lyb]

To me, it is intuitive from last year's spreadsheet that none of the applicants with evaluation to ranking changed on 2nd March or afternoon of 1st March was granted. To be honest, the correlation between the evaluation/ranking changing time and the winning chance is also clear in general. The two things that may challenge these conclusions were the insufficient samples and the uncertainty of the time when they checked the status.

What about the folks that remained in evaluation until the end and still got granted? There's at least one in the spreadsheet, perhaps more that didn't add the information

Yes, but his/her situation was very special. He/She even did not change from submission to evaluation on 3rd March. A possible case maybe that the panel needs to reevaluate his/her Evaluation Report for some reasons. But it indeed brings some hope.

[CGN]

To me, it is intuitive from last year's spreadsheet that none of the applicants with evaluation to ranking changed on 2nd March or afternoon of 1st March was granted. To be honest, the correlation between the evaluation/ranking changing time and the winning chance is also clear in general. The two things that may challenge these conclusions were the insufficient samples and the uncertainty of the time when they checked the status.

I agree, in general it seems that those who changed the first day got the grant or were in reseve list, the following day there were seals of excellence, and then lower scores. But as far as I know, other years went differently.

[Abz]

Last year mine went into ranking quite quickly after we became aware that proposals were being ranked, and I got a score of 99. The first person who realised his proposal went into ranking and alerted us got 100. It was different panels, mine was SOC and his was CHE.

Sometimes they use a different approach and people who aren't ranked early get the fellowship. But generally, the quicker the change to ranking the higher the score.

[lyb]

To me, it is intuitive from last year's spreadsheet that none of the applicants with evaluation to ranking changed on 2nd March or afternoon of 1st March was granted. To be honest, the correlation between the evaluation/ranking changing time and the winning chance is also clear in general. The two things that may challenge these conclusions were the insufficient samples and the uncertainty of the time when they checked the status.

I agree, in general it seems that those who changed the first day got the grant or were in reseve list, the following day there were seals of excellence, and then lower scores. But as far as I know, other years went differently.

I actually checked this Forum in previous years with the key word 'ranking'. In general, it always follows this trend. However, there was one special year when the changing to Ranking happened twice (2017 or 2018, not sure), and the 2nd ranked batch got awarded. But it has never happened again so far.
Things can always change this year. And this is the meaning of filling out this spreadsheet every year, to find a general scenario for the applicants in next years.

[CGN]

And this year, the first people to get ranked were on the 31st of January. I got ranked two days later, so it must be a low score.