Counting observations from a multivariate logistic regression

[Joceynne McRae]

Hello,

I wish to know how I can determine the number of observations for an outcome variable (case or control) by an independent variable (intervention) that has been incorporated in an adjusted logistic regression model for a case-control study.

In this example the model has dropped 7 observations due to "predicting failure perfectly". So I would need to show a total of cases/controls x intervention which also has excluded these seven observations.

I have installed the "distinct" module, but I don't know how to apply it to a regression analysis as opposed to an entire dataset. Perhaps this is the wrong approach.

Any help is appreciated.

Thank you
Jocelynne

logistic outcome independent_1 i.age_mth_test_grp i.family_smoking_miss i.state2 i.month if year==2016 & tested_7d==0 & age_mth_test_grp<5
note: 2.state2 != 0 predicts success perfectly
2.state2 dropped and 4 obs not used

note: 5.state2 != 0 predicts failure perfectly
5.state2 dropped and 3 obs not used


Logistic regression Number of obs = 63
LR chi2(15) = 12.08
Prob > chi2 = 0.6728
Log likelihood = -33.330255 Pseudo R2 = 0.1534

-------------------------------------------------------------------------------------
outcome | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
--------------------+----------------------------------------------------------------
independent_1 | 1.022343 .7859253 0.03 0.977 .2265874 4.61272
|
age_mth_test_grp |
1 | 3.119581 2.807651 1.26 0.206 .5345734 18.20477
2 | .9349802 1.190389 -0.05 0.958 .0771041 11.33776
3 | 2.846745 3.486589 0.85 0.393 .2581238 31.39562
4 | 2.495358 2.854672 0.80 0.424 .2650723 23.491
|
|
family_smoking_miss |
1 | 7.982311 9.941517 1.67 0.095 .6950327 91.67524
9 | 4.259453 9.634958 0.64 0.522 .0505733 358.7456
|
state2 |
2 | 1 (empty)
4 | 1.123285 .9685987 0.13 0.893 .2072555 6.087986
5 | 1 (empty)
|
month |
6 | .5933001 1.113463 -0.28 0.781 .0149899 23.48274
7 | .1460062 .2688005 -1.05 0.296 .0039563 5.388294
8 | .3355456 .587782 -0.62 0.533 .0108314 10.39485
9 | .0646431 .129849 -1.36 0.173 .001261 3.313871
10 | .2031119 .4800371 -0.67 0.500 .001977 20.86772
|
_cons | .7504512 1.38808 -0.16 0.877 .0199935 28.16803
-------------------------------------------------------------------------------------

[Maarten Buis]

// open example data
sysuse nlsw88, clear

// prepare the data

gen byte highoc = occupation < 3 if !missing(occupation)
label variable highoc "high occupation"
label define highoc 1 "higher" ///
                    0 "lower"
label value highoc highoc


// estimate logistic regression
logit highoc i.collgrad ttl_exp i.race i.south i.union, or

// find the number of observations by highoc and collgrad
// for those observations used in the previous model
tab highoc collgrad if e(sample)

// find the number of observations by highoc and collgrad
// for those observations in the data
tab highoc collgrad

[Nick Cox]

That is at most multiple logistic regression, where multiple means having several predictors. That term is fast fading away as a term of art: having several predictors is the norm, not exceptional.

"Multivariate" as applied to models should mean several variables as responses or outcomes: see e.g. mvreg.

To see what has been omitted (a term I prefer to dropped here, but that really is taste) you can go after the model fit

list outcome independent_1 age_mth_test_grp family_smoking_miss state2 month if year==2016 & tested_7d==0 & age_mth_test_grp<5 & !e(sample) 

where !e(sample) is code for "not in the estimation sample". See e.g. https://www.stata-journal.com/articl...article=dm0030 for more.

I see 15 parameters being estimated from 63 observations, which seems optimistic to me, but you may well be on a road towards a simpler model.

[Joceynne McRae]

Thank you both very kindly, I have been able to get the number breakdown needed.

Very helpful,
Cheers
Jocelynne