You might consider Projection Pursuit Regression (ppr function). Since the ranking is just a monotone transformation of the underlying score, ppr can estimate the transformation and the contribution of the terms into the score.
On Mon, Sep 16, 2013 at 11:53 AM, Saumya Gupta <saumya.gu...@outlook.com>wrote: > I have a training dataset which contains statistics of football players > for the year 2009, and their ranks for the year 2010. For example: > > Player No. of goals No. of matches Age Rank (in 2010) A 5 1 35 1 B 2 > 4 23 2 C 1 7 26 3 D 6 4 52 5 E 3 2 45 4 > > The above ranks have been calculated on the basis of a 'score' (which is > unknown) given to each player, which is a function of the 3 variables. It > could be something like: > Score = (No. of goals/No. of matches) - Age^(1/2) > After you arrange the scores in descending order, you get the ranks. The > scores are not known and neither is the formula for calculating it. Only > the ranks are given. > > Now, I have statistics for football players for the year 2013, and I have > to predict their ranks for the year 2014, which should give the same result > if the formula used in 2010 were used. Calculating their scores is not > necessary and even finding out the formula is not the objective. The > objective is just to predict their ranks. But, finding the exact formula > for calculating scores will be a bonus. > > ------------------------------ > Date: Mon, 16 Sep 2013 10:20:08 -0600 > Subject: Re: [R] Regression model for predicting ranks of the dependent > variable > From: 538...@gmail.com > To: saumya.gu...@outlook.com > CC: r-help@r-project.org > > > What question (or questions) are you trying to answer? Any advice we may > give will depend on what you are trying to accomplish. > > > On Sat, Sep 14, 2013 at 2:12 PM, Saumya Gupta <saumya.gu...@outlook.com>wrote: > > I have a dataset which has several predictor variables and a dependent > variable, "score" (which is numeric). The score for each row is calculated > using a formula which uses some of the predictor variables. But, the > "score" figures are not explicitly given in the dataset. The scores are > only arranged in ascending order, and the ranks of the numbers are given > (like 1, 2, 3, 4, etc.; rank 1 means that the particular row had the > highest score, 2 means it had the second highest score and so on). So, if > the data has 100 rows, the output has ranks from 1 to 100. > I don't think it would be proper to treat the output column as a numeric > one, since it is an ordinal variable, and the distance (difference in > scores) between ranks 1 and 2 may not be the same as that between ranks 2 > and 3. However, most R regression models for ordinal regression are made > for output such as (high, medium, low), where each level of the output does > not necessarily correspond to a unique row. In my case, each output (rank) > corresponds to a unique row. > So please suggest me what models I could use for this problem. Will > treating the output as numeric instead of ordinal be a reasonable > approximation? Or will the usual models for ordinal regression work on this > dataset as well? > [[alternative HTML version deleted]] > > ______________________________________________ > R-help@r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. > > > > > -- > Gregory (Greg) L. Snow Ph.D. > 538...@gmail.com > -- Gregory (Greg) L. Snow Ph.D. 538...@gmail.com [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
