Error of "invalid subcommand cv" after estat cv

[Alex Mai]

Hello,

I would like to calculate the coefficient of variation for a give variable. So I run -mean variable- and -estat cv-.
However, then Stata reports "invalid subcommand cv" (r(321)).
If I run -estat sd-, then it works.
So why Stata gave such an alert for estat cv? and what shall I do for this problem?

Thank you very much.

[Clyde Schechter]

Well, the surprise is not that -estat cv- didn't work, but that -estat sd- did! The help for mean postestimation shows that the only thing you can get from -estat- following -mean- is estat vce. You can get a coefficient of variation with -estat cv- after -svy: mean-. To do that, however, you must first -svyset- your data.

[Carlo Lazzaro]

Alex:
the cv of a given variable is the ratio between its standard deviation and mean (usually expressed in pecentage terms).
If you do not have the chance to -svy- your data, what follows will do the trick:

. sysuse auto.dta
(1978 Automobile Data)

. quietly sum price

. di (r(sd)/r(mean))*100
47.840601

In this toy-example, the cv is about 47.48%

[Alex Mai]

Alex:
the cv of a given variable is the ratio between its standard deviation and mean (usually expressed in pecentage terms).
If you do not have the chance to -svy- your data, what follows will do the trick:

. sysuse auto.dta
(1978 Automobile Data)

. quietly sum price

. di (r(sd)/r(mean))*100
47.840601

In this toy-example, the cv is about 47.48%

Dear Clyde and Carlo,
Thank you so much! Actually, when I run -estat cv- after svyset with adjustment of cluster (psu) and stratum (strata), cv cannot be calculated, because STATA reports "Missing standard error because of stratum with single sampling unit".

If I do not adjust cluster and stratum but only take into account sampling weight (pweight), then cv can be reported.

But I am not sure if such a cv without adjusting cluster and stratum is reliable.

Thank you again!

[Carlo Lazzaro]

Alex:
no, you cannot ignore the survey structure of your data.
My previous toy-example assumed that your data were not structured that way.

[Alex Mai]

Alex:
no, you cannot ignore the survey structure of your data.
My previous toy-example assumed that your data were not structured that way.

Dear Carlo,

Thank you! So does it mean that with such a survey database (with singleton strata) it is not possible to calculate correct CV?
In an answer to a previous post, Clyde suggested to re-assign singleton strata to other strata, but this may not be easy (because of the choice of a suitable target stratum, and the large number of singleton strata in my database).

[Carlo Lazzaro]

Alex:
I'm not an expert in survey data analysis.
Soory I cannot be more helpful.

[Alex Mai]

Alex:
I'm not an expert in survey data analysis.
Soory I cannot be more helpful.

Thank you!

[Clyde Schechter]

Thank you! So does it mean that with such a survey database (with singleton strata) it is not possible to calculate correct CV?
In an answer to a previous post, Clyde suggested to re-assign singleton strata to other strata, but this may not be easy (because of the choice of a suitable target stratum, and the large number of singleton strata in my database).

As far as I know, there is no way around this.

[Alex Mai]

As far as I know, there is no way around this.

Dear Clyde,
Thank you! I just now read from other posts about the option of singleunit after -strata()- (it is recommended to use singleunit (centered)). I tried this and it indeed worked for producing CV for database with singleton stratum.
Is this a good method to solve the singleton problem?

Thank you!

[Clyde Schechter]

I'm not familiar with it, and I don't really know. Perhaps somebody else will answer this.

[Alex Mai]

I'm not familiar with it, and I don't really know. Perhaps somebody else will answer this.

Thank you!