Median from multiple variables

[Nikki Ponce]

Hi everyone,

After hours of searching, I cannot figure out how to take the median of multiple columns.
My attempts:

1. create a local list that saved all the values of multiple variables, then tried to find the median of the list
2. create a new variable with the medians for each row; then, took the median of that column

Is there another way to find the median from multiple columns/variables?

Thank you all so much!

[Clyde Schechter]

-help egen- and look at the -rowmedian()- function.

[Nikki Ponce]

Hi Mr. Schechter,

That was my first attempt: generated a column of medians using -rowmedian()- function; then, took the median of that column.
My question is a little confusing - I want to get one median from multiple columns. I guess my first attempt would/should produce the same median if I manually took all the values from multiple columns and took the median of this?

[Nick Cox]

No; the median of a matrix is not necessarily equal to the median of its row or column vectors. Consider 1, 98, 99; 1, 42, 99; 1, 1, 1. The median of the medians is 42 but the overall median is 1. Often you will get closer than that, but there are no guarantees.

This worked for me after installing moremata from SSC.

clear
input y1 y2 y3
1  1 1
1 42 43
1 98 99
end

mata
data = st_data(., "y1 y2 y3")
mm_median(vec(data))

  1

Another method is to reshape long your data first.

[Nikki Ponce]

Hi Professor Cox,

I dissected your solution:

ssc install moremata

mata
data = st_data(., "y1 y2 y3", 0)
mm_median(vec(data))
end

st_data()

creates a sub dataset of the columns I want to take the median of/combine
., returns all observations
I added selectvar==0 to exclude missing values

vec()

stacks multiple columns into one column

mm_median()

uses median function from -moremata- package

Question: If vec() transforms the matrix into one column, why can't we treat the resulting column as a regular column? I use can't because I've tried and failed to get any result when I performed calculations using the vector.

[Nick Cox]

The effect of

vec()

is not permanent any more than y = x + 2 would change x by adding 2 to it.

If I understand you correctly you need an assignment, say

wanted = vec(data)

[Nick Cox]

Just to note that #4 was not well phrased: the key sentence should be

No; the median of a matrix is not necessarily equal to the median of the medians of its row or column vectors.

[Nikki Ponce]

Hi Professor Cox,

The median function worked so I don't necessarily need an assignment. I was just wondering why this won't work:

wanted = vec(data)
summarize wanted, detail

I get an invalid expression error but I think your note about vec() addresses this.

[Nick Cox]

That’s mixes Mata and Stata. wanted is not a Stata variable; you can’t feed a Mata vector directly to a command like summarize.

[Nikki Ponce]

Thank you, Professors Schechter and Cox!