Conditional Group Means

[Lucas Alagoas]

Hi, Suppose, I have the following dataset:

ID	GrouID	X1	X2
1	1	7	10
2	1	7	100
3	1	11	200
4	1	11	100
5	1	50	10

I want to generate three variable as follows:

• Variable 1: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X2_i = X2_j. Example: for ID=1, the new variable would take a value of 100.
• Variable 2: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X2_i < X2_j. Example: for ID=1, the new variable would take a value of (200+100+10)/3.
• Variable 3: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X2_i < X2_j. Example: for ID=1, the new variable would take a value of (10+100+200+100)/4.

I appreciate your help a lot!

[Nick Cox]

Sorry, but I can only understand this by changing all the definitions! In particular you write X2 sometimes where you mean X1 and I think your second inequality should be >= not >.

Note that the text for the second two definitions is identical, so there is some error there.

Variable 1: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1_i = X1_j. Example: for ID=1, the new variable would take a value of 100.

Variable 2: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1_i > X1_j. Example: for ID=1, the new variable would take a value of (200+100+10)/3.

Variable 3: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1_i >= X1_j. Example: for ID=1, the new variable would take a value of (10+100+200+100)/4.

Your data example can be used but in general we mean what we ask in asking that you use dataex (SSC) to show data.

With rangestat (SSC) I get these results Here the intervals 1 1000 and 0 1000 are arbitrary ways to get values in the right intervals. You might need to change 1 or 1000.

clear
input ID    GroupID    X1    X2
1    1    7    10
2    1    7    100
3    1    11    200
4    1    11    100
5    1    50    10
end

rangestat mean1 = X2, interval(X1 0 0) excludeself by(GroupID)

rangestat mean2 = X2, interval(X1 1 1000) excludeself by(GroupID)

rangestat mean3 = X2, interval(X1 0 1000) excludeself by(GroupID)

list

     +-----------------------------------------------------+
     | ID   GroupID   X1    X2   mean1       mean2   mean3 |
     |-----------------------------------------------------|
  1. |  1         1    7    10     100   103.33333   102.5 |
  2. |  2         1    7   100      10   103.33333      80 |
  3. |  3         1   11   200     100          10      55 |
  4. |  4         1   11   100     200          10     105 |
  5. |  5         1   50    10       .           .       . |
     +-----------------------------------------------------+

For the original announcement of this program see http://www.statalist.org/forums/foru...s-within-range

Search the forum for other mentions.

[Nick Cox]

Two of the definitions are still wrong. That's me to blame too.

In this version I have made the notation more Stata-like.

Variable 1: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1[j] = X1[i]. Example: for ID=1, the new variable would take a value of 100.

Variable 2: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1[j] > X1[i]. Example: for ID=1, the new variable would take a value of (200+100+10)/3.

Variable 3: For each individual i, the value of this variable is the average over X2 for all j≠i, given that X1[j] >= X1[i]. Example: for ID=1, the new variable would take a value of (10+100+200+100)/4.

The examples are crucially helpful in indicating what is wanted.

[Lucas Alagoas]

Sorry for messing up all the definitions and for not using dataex (unfortunately, I don't know how to edit my initial post)! What you provided is exactly what I was looking for: Thank you very much!

[Nick Cox]

Thanks for the closure.

Editing the original post is only possible within 1 hour of first posting. A good reason for that restriction is that such editing might make nonsense of much of the following discussion.