Why is my PROC MEANS (w/ class statement) generating combinations that indicate missing values for some of my variables when no missing values exist?

[mconover]

I am attempting to write a program using PROC MEANS that outputs the median, 25th percentile, 75th percentiles, and sum for different combinations of values for 6 variables: cat, ser, fun, caf, lud, wer.

In order to do so I've written the following program, which seems to work fairly well:

PROC MEANS NOPRINT data=work.DATA_collapsed;

     CLASS cat ser fun caf lud wer;

     VAR SUMPay;

     BY SEX;

     OUTPUT OUT=work.PricesForCombos median=med p25=Q1 p75=Q3 sum=SUMPay;

RUN;

PROC PRINT DATA=work.PricesForCombos;

RUN;

Unfortunately, my output shows combinations for my variables in which they have missing values (see output at bottom).  However, none of my variables actually have missing values (values only = 0 and 1).  I confirmed this with the following PROC FREQ:

PROC FREQ DATA=work.DATA_collapsed;

  TABLES cat ser fun caf lud wer /missprint;

RUN;

Is there any simple explanation for this?  What am I missing?_

Output (sorry that the header isn't aligned but hopefully you get the idea)

Obs SEX cat ser fun caf lud wer  _TYPE_  _FREQ_    med      Q1    Q3       SUMPay

  875   2    1    1   0    .    1    .     58     1258    697.25  463.500   956.79   982322.25

  876   2    1    1   1    .    0    .     58      454    712.33  453.600   983.00   359492.67

  877   2    1    1   1    .    1    .     58     2000    844.70  600.525  1113.07  1850175.81

  878   2    1    0   0    .    0    0     59      570    251.17  131.720   355.94   154409.88

  879   2    1    0   0    .    0    1     59       11    162.65  116.800   360.00     3923.75

  880   2    1    0   0    .    1    0     59      157    395.12  217.000   587.20    74664.93

  881   2    1    0   0    .    1    1     59        9    439.00  209.820   903.50     6667.98

  882   2    1    0   1    .    0    0     59       91    460.87  302.780   617.39    47802.26

  883   2    1    0   1    .    0    1     59        2    217.87  216.240   219.50      435.74

  884   2    1    0   1    .    1    0     59      154    495.40  254.000   630.34    81060.69

  885   2    1    0   1    .    1    1     59        6    606.79  508.040   751.00     4950.81

  886   2    1    1   0    .    0    0     59      353    500.90  311.430   742.65   202470.20

  887   2    1    1   0    .    0    1     59       33    800.00  427.000  1041.40    25969.93

  888   2    1    1   0    .    1    0     59     1110    677.76  456.250   917.99   837920.36

  889   2    1    1   0    .    1    1     59      148    872.72  529.555  1276.86   144401.89

  890   2    1    1   1    .    0    0     59      444    714.59  463.880   985.80   352264.78

  891   2    1    1   1    .    0    1     59       10    559.26  277.280   917.96     7227.90

  892   2    1    1   1    .    1    0     59     1937    843.15  600.230  1101.56  1760746.16

  893   2    1    1   1    .    1    1     59       63    900.79  602.120  1533.20    89429.65

  894   2    1    0   0    0    .    .     60      677    257.00  133.590   376.53   195024.39

  895   2    1    0   0    1    .    .     60       70    457.98  282.500   725.23    44642.15

  896   2    1    0   1    0    .    .     60      175    460.00  267.070   609.88    86035.06

  897   2    1    0   1    1    .    .     60       78    523.75  325.730   682.50    48214.44

[Tom · Accepted solution]

The _TYPE_ variable tells you which of the CLASS variables have been used.  For example 58 is '111010'B so the first, second, third and fifth class variables were used to form the combinations.  The fourth and sixth class variables were ignored, hence the missing values for CAF and WER.

If you do NOT want it to generate the summaries for all combinations of the class variables then add the NWAY option to the PROC MEANS statement.

[mconover]

I see.  That makes sense.  I really appreciate your help. 

Just to confirm, when I'm seeing "." in my output, it's not actually referring to those values being missing, but rather that that variable is not a component of the displayed combination.  Correct?

[Tom]

In your case, yes, since you did not include the MISSING option on the PROC MEANS statement.  Without that option any observation that had a missing value for any of the class variables would have been excluded from the processing.