Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- Base SAS Programming
- /
- Ranking

- RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

06-07-2017 05:57 AM

Hello dear SAS community.

I have a real struggle at the moment, and I'm running out of time.

To make things short: I want for each year, rank all firms by X for that year. But only max "5 Ranks" fo reach year, and start in year 1968 for example, and repeat it each year.

For example:

FirmID YEAR VAR_X WANT

1000 1965 9 3 4 (69)

1000 1966 10 2 3(69)

1000 1967 11 1 2(69)

1000 1968 12 1(69)

1000 1969 16

1000 1970 19

1000 1971 20

1000 1972 21

.

XXXX XXXX XXXX XXXX

BUT the problem starts when I'm doing this in year 1969, since it contains a "rank" variable in pre 68's.

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to TorTheHammer

06-07-2017 10:50 AM

What precisely does the WANT value [ 3 4 (60) ] mean in the data line for 1965.

FirmID YEAR VAR_X WANT

1000 1965 9 3 4 (69)

It appears you want, for each year, to

- Rank the preceding 4 ( preceding 5?) years. I notice for the "(69)" you don't actually assign a rank value to the 1969 data.
- Assign those ranks to those preceding years.

This means you would want, for each year t, 4 RANK vars

- RANK_0 = rank of X(t) among x{t-3}, x{t-2}, x{t-1}, x{t}
- RANK_1 = rank of X{t} among x{t-2} x{t-1}, x{t}, x{t+1}
- ...
- RANK_3 = rank of x{t} among x{t}, t{t+1}, x{t+2}, x{t+3};

Is that what you mean?

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to mkeintz

06-07-2017 01:31 PM

Also, what rank do you want to assign when there are ties in the x values?

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to mkeintz

06-07-2017 06:36 PM

Yes, this is what I want. Sorry for bad "writing" about it

Sorry for asking rude, but can you help me about it?

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to TorTheHammer

06-07-2017 10:51 AM

I think it may help to provide a little more raw input data and how you want the result at the end of the process. You say " for each year" but your example only has one year of ranks (apparently).

Also please explain where you got your "want" values from. If 3 is supposed to be Rank, how did you get 3? The value of 9 for VAR_x is the smallest shown so I would expect it to be either 1 (if you want to rank or increasing values of var_x) or not ranked if looking at decreaseing values. And what is 4?

How do you get 2 values of 1? Are you perhaps missing a bunch of different variables to hold multiple periods of consideration but didn't show them?

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to TorTheHammer

06-07-2017 03:01 PM - edited 06-07-2017 03:07 PM

The DATA WANT step below does what I think you want. If there are ties among the X's being ranked, then it produces the average of the lowest and highest rank assigned to the X value:

It works by:

- Reading in each record for a gien FIRMID and storing X in an array, indexed on year, with a lowest bound 1960 and highest bound of 2010. This assumes that the lowest observable year is 1963 and the highest observed year is 2010.
- When the end of a firmid is encountered, then, in a loop, reread each year, and capture the rank of its X value versus (1) years YEAR-3 through YEAR for RANK_), (2) years YEAR-2 through YEAR+1 for RANK_1, etc.

```
data have;
input FirmID YEAR X;
output;
firmid=2000; if mod(_n_,2)=0 then x=x-2; output;
datalines;
1000 1965 9
1000 1966 10
1000 1967 11
1000 1968 12
1000 1969 16
1000 1970 19
1000 1971 20
1000 1972 21
run;
proc sort; by firmid year;run;
data want (drop=_:);
array _x{1960:2010} _temporary_;
set have (keep=firmid year x );
by firmid;
if first.firmid then call missing(of _x{*});
_x{year}=x; /* Fill the _X array */
label rank_0 = "Rank of current X for Current Year-3 through Current Year"
rank_1 = "Rank of current X for Current Year-2 through Current Year+1"
rank_2 = "Rank of current X for Current Year-1 through Current Year+2"
rank_3 = "Rank of current X for Current Year through Current Year+3" ;
array rnk{0:3} rank_0-rank_3;
if last.firmid then do until (last.firmid);
set have ;
by firmid;
do _r=0 to 3;
do _rank_beg=1 by 1 until (largest(_rank_beg ,_x{year+_r-3},_x{year+_r-2},_x{year+_r-1},_x{year+_r-0})=x);end;
if _rank_beg<4 then do _rank_end=_rank_beg to 3 while (largest(_rank_end+1,_x{year+_r-3},_x{year+_r-2},_x{year+_r-1},_x{year+_r-0})=x);end;
else _rank_end=4;
rnk{_r}=(_rank_beg+_rank_end)/2;
end;
output;
end;
run;
```

Editted addition. To deal with possible ties for X, you see two consecutive loops

DO _RANK_BEG= 1 by 1 until (.....);

IF _RANK_BEG<4 then DO _RANK_END=_RANK_BEG to 3 WHILE (.....);

If you expect no ties in X, you could replace them with a single loop:

DO RNK{_r}=**1** by **1** until (largest(rnk{_r},_x{year+_r-**3**},_x{year+_r-**2**},_x{year+_r-**1**},_x{year+_r-**0**})=x);