Re: st: Nearest neighbor distance

[Nick Cox <njcoxstata@gmail.com>]

-nearest- is a user-written program from SSC. You are asked to
identify where user-written programs you refer to come from.

Your problem is both similar to and very different from that of
-nearest- and you would need to rewrite -nearest-. I wouldn't call
that a slight modification.

-nearest- is indifferent to ties and whether nearest neighours are
reflexive, i.e. A is the nearest neighbour of B, and also vice versa.
These could be bigger issues with your kind of data.

I have no idea what the Kogut and Singh index is.

Nick

On Tue, Aug 23, 2011 at 6:27 PM, Lange, Sandra <Sandra.Lange@whu.edu> wrote:
> I would like to modify the code of the stata command 'nearest' to identify the closest neighbor (from a defined set of observations) for specific observations in a panel data set.
> I work with an unbalanced sample of firms which ranges over a time period of about 20 years.
> The dataset contains the portfolio of subsidiaries of each firm in each year and consists of over 100,000 observations (one observation = subsidiary of a firm in one year). In addition, several country characteristics were merged into the dataset. Below you find an excerpt to get an impression of how the data looks like:
> firm_id unit_id year    status  country countryname     pdi     idv     mas     uai     subyears        nearest nearest_id
> 100     15      1990    U       215     Japan           54      46      95      92          2
> 100     44      1990    I       235     Russia          93      39      36      95          0
> 100     4       1990    U       404     Belgium 65      75      54      94      3
> 100     46      1990    I       408     Germany 35      67      66      65      0
> 100     18      1990    U       408     Germany 35      67      66      65      4
> 100     2       1990    U       408     Germany 35      67      66      65      4
> 100     38      1990    I       434     Switzerland     34      68      70      58      0
> 100     15      1991    U       215     Japan           54      46      95      92      3
> 100     44      1991    U       235     Russia          93      39      36      95      1
> 100     4       1991    U       404     Belgium 65      75      54      94      4
> 100     46      1991    U       408     Germany 35      67      66      65      7
> 100     18      1991    U       408     Germany 35      67      66      65      7
> 100     2       1991    U       408     Germany 35      67      66      65      7
> 100     38      1991    U       434     Switzerland     34      68      70      58      1
> 100     54      1991    I       429     Poland          68      60      64      93      0
> 100     53      1991    I       429     Poland          68      60      64      93      0
> 100     51      1991    I       430     Portugal        63      27      31      104     0
> .       .       .       .       .       .       .       .       .       .       .       ...
> 101     181     1985    U       215     Japan           54      46      95      92      1
> 101     150     1985    U       236     Saudi-Arabia    80      38      52      68      1
> 101     146     1985    U       237     Singapur        74      20      48      8       1
> 101     140     1985    U       404     Belgium 65      75      54      94      2
> 101     155     1985    U       408     Germany 35      67      66      65      3
> 101     83      1985    U       408     Germany 35      67      66      65      3
> 101     84      1985    U       408     Germany 35      67      66      65      3
> 101     133     1985    U       411     France          68      71      43      86      2
> 101     147     1985    U       411     France          68      71      43      86      2
> 101     222     1985    I       438     Spain           34      51      42      86      0
> .       .       .       .       .       .       .       .       .       .       .
>
> More precisely, this is what I would like to do:
>
> 1.  for each observation with status 'I' (Investment), I am looking for the closest country in terms of cultural dimensions (pdi, idv, mas, uai) in the firms   existing portfolio (observations with status 'U'). I suppose I could use the code for 'nearest'; however, I probably would have to change it slightly, because the 'nearest' command finds the closest neighbor in N; however, I am looking for the closest neighbor in _n which should be somehow specified as the existing portfolio (all subsidiary-year observations with status ==U).
> - Is it possible to modify the code of the command 'nearest' for that in the first place? Does someone have a suggestion?
> - How should I deal with the fact that I have multiple dimensions in the code of the command 'nearest'? I want to use the Kogut&Singh index for calculating
>  the distance based on these four dimensions. At some point I would have to indicate that, but I do not know where.
>
> 2. A slight modification of 1.: for each observation with status 'I' (Investment), I am looking for the closest country (in the firms existing portfolio)  in terms of cultural dimensions (pdi, idv, mas, uai) AND subyears. If subyears < 5, then the country should not qualify for being selected as the closest neighbor. In this case the second closest neighbor should be chosen and checked if subyears >= 5. Otherwise the third closest neighbor should be investigated, and so on.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/