This is an example of PROC SGPANEL:   proc sgpanel data=sashelp.prdsale; where product in ("CHAIR" "SOFA"); title "Yearly Sales by Product"; panelby year/spacing=5 novarname rows=2 columns=1; vline month/response=actual group=product markers; run;   The output graph prints year (1993, 1994) as indicator for each panel. How can I not print it (and where will the option be put in the program)?   I'm asking this in cases where the purpose of dividing data into panel is because there will be too many lines to pack in a graph, i.e., the variable used for such division has no actual meaning.     ... View more

Several continuous variables in my data have outliers/extreme values. Instead of assigning them missing values, I wonder if there's a more non-invasive approach, i.e., to leave outliers untouched but run statistics that exclude them. For example, how can I run means of a variable that excludes 5 observations with lowest values, and maybe 5 others highest values? ... View more

Consider the hypothetical setup: X denotes counties in a city. We can have X1, X2,..., Xi. Y denotes hospitals in the city. So let's say for county X1, we have Y11, Y12,..., Y1j, and so on for other counties. Z denotes number of nurses in these hospitals. So for county X1, we have Z11 for hospital Y11, Z12 for hospital Y12,... Z1j for hospital Y1j, and so on. We take a sample of hospitals, so not all of them are included in the sample. Let's say: x denotes counties in the sample, x1, x2,..., x(i) y denotes hospitals in the sample. For x1 we may have y1, y2,..., y(u), and so on. z denotes number of nurses in the sample's hospitals. For x1 we have z1, z2,..., z(u). I want to use the data from the sample to estimate the number of nurses in the target population (the city). Using mean value in a quick and dirty fashion for this purpose, I think of two potential approaches: In the first approach, simply assuming the overall mean of nurse in the sample represents the mean of population. That leads to the following estimation of Z: Z = (mean of z)*Y In the second approach, I will calculate in a similar way but for each county and summing up those individual estimates. So for county X1, Z1 = (z1 + z2 + ... + z(u))/(number of sampled hospitals of county x1)*number of hospitals of county X1 After that: Z = Z1 + Z2 + ... + Zi   It turns out that the two methods produce different results most of the time. I wonder which is the better estimates. ... View more

The input data looks like below:    State County HH Var 1 1 1 1 1 1 2 2 1 1 3   1 2 1 4 1 2 2 5 1 3 1 6 2 1 1 1 2 1 2 2 2 2 1 3 2 3 1 4 2 3 2     The set of 3 variables state, county, and households (HH) specifies unique codes for observations.   The variable Var is, say, a characteristic of households and it has missing data. My question is: How can I replace those missing observations with the mean values of households of the same county?   The output data should look like:   State County HH Var 1 1 1 1 1 1 2 2 1 1 3 1.5 1 2 1 4 1 2 2 5 1 3 1 6 2 1 1 1 2 1 2 2 2 2 1 3 2 3 1 4 2 3 2 4     ... View more

The title sounds vague, but here the data looks like: data temp; input ID NumberOfMembers Mem1Var1 Mem1Var2 Mem2Var1 Mem2Var2 Mem3Var1 Mem3Var2; datalines; ID01     1     1     0     .     .     .     . ID02     2     1     .     0     1     .     . ID03     3     0    1     1     0     0     0 ; ID represents a type of group variable, say, families. Variables from Mem1Var1 to Mem3Var2 are characteristics measured for individual members in each group. Missing values are assigned if there are not enough members of a group to fill in these characteristics. E.g., group ID01 consists of only 1 member, thus values for Mem2Var1, Mem2Var2, Mem3Var1 and Mem3Var2 are all 0s. Let's call these "mechanical missings". There are other missing values due to data collection etc., such as Mem1Var2 of ID02. Let's call these "other missings". Now how can I assign the "mechanical missings" with a specific value, let's say 9? So the expected data looks like: ID01     1     1     0     9     9     9     9 ID02     2     1     1     0     1     9     9 ID03     3     .     1     1     0     0     0 If it matters, the reason for doing so is to distinguish this type of missings. They are not really "missing" as a consequence of, for example, an error in data collection or entry leading to the incompleteness of data. Instead, they don't have values because there are no observations for those values to be measured. I think this distinguishing is useful is some cases. For example, if you transpose the data to the long form by variable 2, it looks like this for ID02: ID         Member          Var2          NumberOfMembers ID02     1                         .                    2 ID02     2                         1                    2 ID02     3                         .                    2 So assigning specific values will distinguish difference of missing values in Var2 and the description of data will be more accurate. Actually, as I'm typing this, I think we can assign mechanical missings based on long form data: IF NumberOfMembers < Member THEN Var2 = 9; Not sure what to do with wide form though. ... View more

The original data looks like: data have; input Var1 Var2 Var3 Var4; datalines; 1     0     .     0 1     1     1     2 1     .     .     3 0     2     3     1 ; I want to create a data that looks like: 1     0     .     0 1     0     0     0 1     .     .     0 0     2     3     1 ; The idea is if Var1 = 1 then replace all non-missing values of Var2, Var3 and Var4 to 0 and keep missing values as-is. EDIT: Will this work? data want; set have; array varlist Var 2 Var3 Var4; do i = 1 to 4; if Var1 = 1 and varlist(i) ne . then varlist(i) = 0; end; drop i; run; EDIT 2: It turns out to work. Sorry for bringing it up. ... View more

I want to run proc means on several continuous variables, for example Var1, Var2, Var3, but want to leave out certain observations because their values are not appropriate for analysis. For example, some observations/respondents may receive code of 99999999 on any, some, or all of these variables if those respondents were not asked such questions on the survey, like men generally aren't asked how much they spend each month on lotions (well, some surveys may ask that question and I've got no problem with that, but just to say). It can also be the case where there are more than one values that we wish to leave out. For example, some observations receive values of 88888888 if they answered "Don't know". So these are not missing values as they did answer, but at the same time not legitimate for calculating (means) either. If it's one variable, e.g., Var1, then I can do: proc means data=have; where Var1 not in (88888888, 99999999); var Var1; run; But I'm not sure how to make it efficient in the case of multiple variables. I think of a two-step solution but then figure out it won't work, like: data have; set have; Var4=0; array myarr Var1 Var2 Var3; do i = 1 to 3; if myarr(i) in (88888888, 99999999) then Var4=1; end; drop i; run; proc means data=have; where Var4 ne 1; var Var1 Var2 Var3; run; It won't work because the array assigns 1 to Var4 if any from Var1 - Var3 receives one of these values, thus conditioning on Var4 inappropriately leaves out legitimate observations of variables. ... View more