I am working in two locations, in one I am using Stata 13 and in the other Stata 14.
Can I build a do-file that works in both versions even if some specific command has changed?
For instance, the following code will not work using Stata 13
sysuse auto, clear
ci means mpg price, level(90)
but this one works
sysuse auto, clear
ci mpg price, level(90)
Uising Stata 14, it will be the opposite.
I thought about adding capture but nothing happens in Stata 13 or Stata 14.
. sysuse auto, clear
(1978 Automobile Data)
. capture ci means mpg price, level(90)
. capture ci mpg price, level(90)
Update: Adding noisily after capture didn't help unfortunately. Here is an example with Stata 14 that works
. sysuse auto, clear
(1978 Automobile Data)
. capture noisily ci mpg price, level(90)
you must specify one of means, proportions, or variances following ci
. capture noisily ci means mpg price, level(90)
Variable | Obs Mean Std. Err. [90% Conf. Interval]
-------------+---------------------------------------------------------------
mpg | 74 21.2973 .6725511 20.17683 22.41776
price | 74 6165.257 342.8719 5594.033 6736.48
. gen lb=r(lb)
. su lb
Variable | Obs Mean Std. Dev. Min Max
-------------+---------------------------------------------------------
lb | 74 5594.033 0 5594.033 5594.033
But this one does not work when you invert two lines of code (with Stata 14):
. sysuse auto, clear
(1978 Automobile Data)
. capture noisily ci means mpg price, level(90)
Variable | Obs Mean Std. Err. [90% Conf. Interval]
-------------+---------------------------------------------------------------
mpg | 74 21.2973 .6725511 20.17683 22.41776
price | 74 6165.257 342.8719 5594.033 6736.48
. capture noisily ci mpg price, level(90)
you must specify one of means, proportions, or variances following ci
* The program didn't stop but:
. gen lb=r(lb)
(74 missing values generated)
. su lb
Variable | Obs Mean Std. Dev. Min Max
-------------+---------------------------------------------------------
lb | 0
Finally, note that the first code that works correctly with Stata 14 doesn't work with Stata 13
. sysuse auto, clear
(1978 Automobile Data)
. capture noisily ci mpg price, level(90)
Variable | Obs Mean Std. Err. [90% Conf. Interval]
-------------+---------------------------------------------------------------
mpg | 74 21.2973 .6725511 20.17683 22.41776
price | 74 6165.257 342.8719 5594.033 6736.48
. capture noisily ci means mpg price, level(90)
variable means not found
. gen lb=r(lb)
(74 missing values generated)
. su lb
Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
lb | 0
If you wish to use capture to catch an error, there is follow-through as well. Here you first try version 14 syntax, and if and only if that fails you try version 13 syntax.
sysuse auto, clear
capture noisily ci means mpg price, level(90)
if _rc ci mpg price, level(90)
gen lb = r(lb)
Here if _rc is a Stataish abbreviation for if _rc > 0 which will happen if and only if a program fails. _rc of 0 means everything was legal (with minute qualifications). _rc is the return code.
I am not clear that putting a single value in a variable is a good idea, but let that be a different issue. Also, you asked for two confidence intervals, and only results for the first will remain in memory.
Related
I currently struggle to apply the reverse option to my ttest.
Here is a toy example:
sysuse auto,clear
local varlist mpg price weight
eststo foreign: quietly estpost summarize `varlist' if foreign==0
eststo domestic: quietly estpost summarize `varlist' if foreign==1
eststo diff: quietly estpost ttest mpg, by(foreign) unequal **reverse**
esttab foreign domestic diff
The following works:
sysuse auto,clear
local varlist mpg price weight
eststo foreign: quietly estpost summarize `varlist' if foreign==0
eststo domestic: quietly estpost summarize `varlist' if foreign==1
eststo diff: quietly estpost ttest mpg, by(foreign) unequal reverse
esttab foreign domestic diff
Note:
ttest mpg, by(foreign) unequal reverse
works.
After reading the estpost documentation, it seems the package currently does not support the option.
Eventually, I need to create a table of ttests for about 20 variables and reverse the result for b/t. I'm vey thankful for workarounds!
You can do everything using regression like this (even in small samples):
. sysuse auto, clear
(1978 automobile data)
. /* reverse t-test */
. ttest mpg, by(foreign) unequal reverse
Two-sample t test with unequal variances
------------------------------------------------------------------------------
Group | Obs Mean Std. err. Std. dev. [95% conf. interval]
---------+--------------------------------------------------------------------
Foreign | 22 24.77273 1.40951 6.611187 21.84149 27.70396
Domestic | 52 19.82692 .657777 4.743297 18.50638 21.14747
---------+--------------------------------------------------------------------
Combined | 74 21.2973 .6725511 5.785503 19.9569 22.63769
---------+--------------------------------------------------------------------
diff | 4.945804 1.555438 1.771556 8.120053
------------------------------------------------------------------------------
diff = mean(Foreign) - mean(Domestic) t = 3.1797
H0: diff = 0 Satterthwaite's degrees of freedom = 30.5463
Ha: diff < 0 Ha: diff != 0 Ha: diff > 0
Pr(T < t) = 0.9983 Pr(|T| > |t|) = 0.0034 Pr(T > t) = 0.0017
. /* calculate means */
. eststo means: qui reg mpg ibn.foreign, vce(hc2) nocons
. /* regression equivalent using robust veriance with a bias correction and t-test DoF */
. eststo diff: qui margins, dydx(foreign) df(`=scalar(df_t)') post
. esttab means diff, label se
----------------------------------------------------
(1) (2)
Mileage (m~)
----------------------------------------------------
Domestic 19.83*** 0
(0.658) (.)
Foreign 24.77*** 4.946**
(1.410) (1.555)
----------------------------------------------------
Observations 74 74
----------------------------------------------------
Standard errors in parentheses
* p<0.05, ** p<0.01, *** p<0.001
I want to export p-value of means and median difference to latex from stata. I tried the code below and it works for p-value of means. However, I dont know how to add the analysis on the difference between medians. Can you please help me with this?
eststo control: quietly estpost summarize a b c if treated == 0
eststo treated: quietly estpost summarize a b c if treated == 1
eststo diff: quietly estpost ttest a b c, by(treated) unequal
esttab using means_medians.tex, replace mlabels("Treated" "Control" "Difference")
cells("mean(pattern(1 1 0) fmt(2)) sd(pattern(1 1 0 ) fmt(3)) b(star pattern(0 0 1) fmt(2))") label
You can use epctile that is implemented as a proper estimation commands and returns e(V) and the estimation table. (Disregard the ugly output in the first part. I wrote it about 10 years ago for a specific project, but it seems like I have never made it pretty enough for the general use.)
. sysuse auto, clear
(1978 Automobile Data)
. epctile mpg, p(50)
Mean estimation Number of obs = 74
--------------------------------------------------------------
| Mean Std. Err. [95% Conf. Interval]
-------------+------------------------------------------------
__000006 | -.027027 .058435 -.1434878 .0894338
--------------------------------------------------------------
Percentile estimation
------------------------------------------------------------------------------
mpg | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
p50 | 20 .75 26.67 0.000 18.53003 21.46997
------------------------------------------------------------------------------
This command is not on SSC, it is on my page, so to install it, follow the prompts in findit epctile.
. sysuse auto
(1978 Automobile Data)
. centile price, centile(25 75)
-- Binom. Interp. --
Variable | Obs Percentile Centile [95% Conf. Interval]
-------------+-------------------------------------------------------------
price | 74 25 4193 4009.467 4501.838
| 75 6378 5798.432 9691.6
. tabstat price, stat(p25 p75)
variable | p25 p75
-------------+--------------------
price | 4195 6342
When making calculations by hand, my answers agree with the centile command, and disagree with the tabstat command (bonus: they also disagree with the sum , detail command).
Where is this discrepancy (25th percentile: 4193 vs 4195, and 75th percentile: 6378 vs 6342) coming from?
I am using Stata 15.1 for Unix.
I am trying to find a way to get the number of levels of a categorical variable as a single number. For example if I have a variable X with 4 levels I need to somehow get that number. If I type levelsof X I get the following 1 2 3 4 but I can't get only number 4 from there. Is there a way to do it using the levelsof or another command?
Various commands will give you the number of distinct values, for any kind of variable. ("Categorical variable" is a statistical concept, rather than a Stata concept.) Perhaps the simplest way to do it for one-off purposes is to ask for a one-way tabulation using tabulate. The number of distinct values is then the number of rows in that table, returned as r(r). Note that (1) you can suppress the table itself (which is useful in a program or do file) and (2) missing values are excluded by default:
. sysuse auto, clear
(1978 Automobile Data)
. qui tab foreign
. ret li
scalars:
r(N) = 74
r(r) = 2
. qui tab rep78
. ret li
scalars:
r(N) = 69
r(r) = 5
. qui tab rep78, missing
. ret li
scalars:
r(N) = 74
r(r) = 6
An extended review of this problem, pitched more generally, is available here. That paper introduces a distinct command. Its uses include direct support for looking at the number of distinct values systematically. search distinct in Stata to find a download source for the most recent version.
. distinct
| Observations
| total distinct
--------------+----------------------------
make | 74 74
price | 74 74
mpg | 74 21
rep78 | 69 5
headroom | 74 8
trunk | 74 18
weight | 74 64
length | 74 47
turn | 74 18
displacement | 74 31
gear_ratio | 74 36
foreign | 74 2
You can look at r(r) after levelsof:
. sysuse auto
(1978 Automobile Data)
. levelsof rep78
1 2 3 4 5
. display "rep78 has " `r(r)' " levels."
rep78 has 5 levels.
I'd like to split a sample according to a specific variable, creating 4 sub-samples each one related to a quartile of the variable's distribution. The aim is to demonstrate that the presence of different levels of this variable influences the outcome of a regression, making it significant or not.
The easiest way to do this is to use the egen command to cut your variable into four equally-spaced intervals.
Example:
. sysuse auto, clear
(1978 Automobile Data)
. sum price, detail
Price
-------------------------------------------------------------
Percentiles Smallest
1% 3291 3291
5% 3748 3299
10% 3895 3667 Obs 74
25% 4195 3748 Sum of Wgt. 74
50% 5006.5 Mean 6165.257
Largest Std. Dev. 2949.496
75% 6342 13466
90% 11385 13594 Variance 8699526
95% 13466 14500 Skewness 1.653434
99% 15906 15906 Kurtosis 4.819188
. egen price_cut = cut(price), group(4)
. table price_cut, contents(n price min price max price)
----------------------------------------------
price_cut | N(price) min(price) max(price)
----------+-----------------------------------
0 | 18 3,291 4,187
1 | 19 4,195 4,934
2 | 18 5,079 6,303
3 | 19 6,342 15,906
----------------------------------------------
I hope this helps you.
This is the easiest way you can go about it:
xtile xx=yourvariable, nq(4)
I hope this helps.