Output of 'ci proportions' depends on variable format

[Gabor Mihala]

Why does the result of 'ci proportions x' depend on the format of variable x? Very strange. Using version 15. See below:

. describe x

storage display value
variable name type format label variable label
-------------------------------------------------------------------------------------
x byte %8.0f Example variable

. tab x, missing

Example |
variable | Freq. Percent Cum.
------------+-----------------------------------
0 | 86,913 98.73 98.73
1 | 1,119 1.27 100.00
------------+-----------------------------------
Total | 88,032 100.00

. ci proportions x

-- Binomial Exact --
Variable | Obs Proportion Std. Err. [95% Conf. Interval]
-------------+---------------------------------------------------------------
x | 88,032 0 0 0 0

. format x %10.0g

. ci proportions x

-- Binomial Exact --
Variable | Obs Proportion Std. Err. [95% Conf. Interval]
-------------+---------------------------------------------------------------
x | 88,032 .0127113 .0003776 .0119817 .0134733

[Mike Lacy]

I can't reproduce your results. Here's what I get:

. clear

. // create data
. set obs 88032
number of observations (_N) was 0, now 88,032

. gen byte x = _n <= 1119

. //
. //
. desc x

              storage   display    value
variable name   type    format     label      variable label
------------------------------------------------------------------------------------------------------------------------------------------------------
x               byte    %8.0g                 

. tab x

          x |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |     86,913       98.73       98.73
          1 |      1,119        1.27      100.00
------------+-----------------------------------
      Total |     88,032      100.00

. ci proportions x   // works fine for me

                                                         -- Binomial Exact --
    Variable |        Obs  Proportion    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           x |     88,032    .0127113    .0003776        .0119817    .0134733

[Gabor Mihala]

Thanks Mike. I had %8.0f format to start with.
Sorry about the formatting of my first post - otherwise I've been a Stata user since 2011 so this relatively simple issue is very surprising!

[William Lisowski]

Welcome to Statalist, and thank you for the provocative question.

My example below shows that the format assigned to x is applied by ci to all the values reported for x, other than Obs.

This seems like a questionable decision in some cases, especially for 0/1 variables with a %n.0f format assigned, and a decision that should be more easily overridden with an option to the ci command. Nevertheless, ci appears to have operated this way for some time now, so I suppose it is not an oversight.

Added in edit: the summarize command does not suffer from this issue; it apparently ignores the format assigned to x. I don't see why the ci command does what it does.

. clear

. // create data
. set obs 88032
number of observations (_N) was 0, now 88,032

. gen byte x = _n <= 1119

. tab x

          x |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |     86,913       98.73       98.73
          1 |      1,119        1.27      100.00
------------+-----------------------------------
      Total |     88,032      100.00

. // try different formats
. format x %8.0g

. ci proportions x

                                                         -- Binomial Exact --
    Variable |        Obs  Proportion    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           x |     88,032    .0127113    .0003776        .0119817    .0134733

. format x %8.4f

. ci proportions x

                                                         -- Binomial Exact --
    Variable |        Obs  Proportion    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           x |     88,032      0.0127      0.0004          0.0120      0.0135

. format x %8.0f

. ci proportions x

                                                         -- Binomial Exact --
    Variable |        Obs  Proportion    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           x |     88,032           0           0               0           0

. // try older version
. version 13: ci x, binomial

                                                         -- Binomial Exact --
    Variable |        Obs        Mean    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           x |     88,032           0           0               0           0

.

[Gabor Mihala]

Thank you, William.

I tend to use the %8.0f variable format, maybe I should switch to the %8.0g format.

[William Lisowski]

In post #4 I gave an incomplete description of how the summarize command formats its output. It chooses a general format as a default, but includes an option to override the default with the variable's assigned format.

This struck me as an elegant solution to providing a useful default format while allowing customization, so I asked Stata Technical Services about the difference in the approaches between the summarize and ci commands.

In response I was told

Thank you for bringing this to our attention. I have passed this along to the relevant developers, and we will release a fix in a future update. We apologize for any inconvenience.

I appreciate that response.

[Gabor Mihala]

Thank you, William