Comparing prevalence

[Siluleko Mkhize]

Hi community,

I wish to compare the prevalence of food security among adult males (AGE_VQ_P>=20). Food security is an ordinal outcome with 3 levels (food secure, at risk of hunger and food insecure) and male==1.
I have tried using a Kruskal-Wallis and I'm not sure if this test is appropriate in this case. Here is the code and outcome I produced:

kwallis Sex if AGE_VQ_P>=20 & Sex==1, by(Hunger_cat)

Kruskal-Wallis equality-of-populations rank test

+--------------------------------------+
| Hunger_cat | Obs | Rank Sum |
|--------------------------+--------+---------------|
| Food secure | 1,840 | 3.61e+06 |
| At risk of hunger | 960 | 1.88e+06 |
| Experience hunger | 1,119 | 2.19e+06 |
+---------------------------------------------------+

chi-squared = -0.000 with 2 d.f.
probability = 1.0000

chi-squared with ties = . with 2 d.f.
probability = 0.0001

and If this test is appropriate, may you please advise on how to run a Tukey post hoc analysis

[Nick Cox]

You posted a version of this at #7 on https://www.statalist.org/forums/for...al-wallis-test

As I understand it, your Hunger_cat is to be regarded as an outcome or response and Sex is a predictor. That being so, your variables are the wrong way round so that a command like

kwallis Hunger_Cat if AGE_VQ_P>=20, by(Sex)

makes more sense to me. Note also that if Sex == 1 restricts analysis to males and makes no sense if the aim is to think about Sex as a predictor, so compare distributions for males and females.

All that said, you clearly have more variables than Sex and so Kruskal-Wallis looks like a dead end compared with say ordered logit or ordered probit.

To get better advice, you would need to tell us more about your data.

[Siluleko Mkhize]

Running the suggested command gives this output:

kwallis Hunger_cat if AGE_VQ_P>=20, by(Sex)

Kruskal-Wallis equality-of-populations rank test

+---------------------------+
| Sex | Obs | Rank Sum |
|--------+-------+----------|
| Male | 3,919 | 1.91e+07 |
| Female | 6,069 | 3.08e+07 |
+---------------------------+

chi-squared = 9.765 with 1 d.f.
probability = 0.0008

chi-squared with ties = 11.293 with 1 d.f.
probability = 0.0008

Is this not a comparison of the proportions of males vs females instead of the previous one I posted where it seemed like a comparison of food security among males?

Ideally, my data looks something like this:

	Total N	Food secure % (95% CI)	At risk of hunger % (95% CI)	At risk of hunger % (95% CI)	P value
Males	3 919	47.0 (45.4 – 48.5)	24.5 (23.2 – 25.9	28.6 (27.2 – 30.0)
Females	6 069	44.2 (42.9 *– 45.4)	24.0 (23.0 – 25.1)	31.8 (30.6 – 33.0)

I wish to compare the proportions of food security status by row. The proportions were all computed in stata. Now I'm just failing to compare them.

[Siluleko Mkhize]

Or maybe showing the dataset may be more helpful:

Sex	Hunger_cat
Male	Food secure
Female	At risk of hunger
Female	Food insecure

I computed the proportions using:

proportion Hunger_cat if AGE_VQ_P>=20 & Sex==1, over(Sex) cformat(%9.3f)
proportion Hunger_cat if AGE_VQ_P>=20 & Sex==2, over(Sex) cformat(%9.3f)

[Carlo Lazzaro]

Siluleko:
can't you translate your data into a -mlogit- framework?:

mlogit Hunger_cat i.Sex <other predictors/controls>

[Siluleko Mkhize]

Carlo,

The mlogit is testing for an association between my predictors and the outcome variable. I want to compare prevalences as shown in the table with percentages and 95%CI.
I feel so lost, I've been asking around and no one seems to know how to go about.

[Carlo Lazzaro]

Siluleko:
you may want to use -prtesti- for each stratum of -food security-.
An example concerning -Food secure- follows:

. prtesti 3919 0.47 6060 .442

Two-sample test of proportions                     x: Number of obs =     3919
                                                   y: Number of obs =     6060
------------------------------------------------------------------------------
             |       Mean   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           x |        .47   .0079726                       .454374     .485626
           y |       .442   .0063796                      .4294963    .4545037
-------------+----------------------------------------------------------------
        diff |       .028   .0102108                      .0079871    .0480129
             |  under Ho:   .0102038     2.74   0.006
------------------------------------------------------------------------------
        diff = prop(x) - prop(y)                                  z =   2.7441
    Ho: diff = 0

    Ha: diff < 0                 Ha: diff != 0                 Ha: diff > 0
 Pr(Z < z) = 0.9970         Pr(|Z| > |z|) = 0.0061          Pr(Z > z) = 0.0030

[Siluleko Mkhize]

Carlo,

Thanks a lot, that worked. So this test only takes two groups, I can't add three proportions?

[Siluleko Mkhize]

Or rather compare the three proportions within the male category and then the female category

[Joseph Coveney]

I wish to compare the proportions of food security status by row.

Try something like the following.

.ÿ
.ÿversionÿ16.0

.ÿ
.ÿclearÿ*

.ÿ
.ÿinputÿbyteÿsexÿint(count0ÿcount1ÿcount2)

ÿÿÿÿÿÿÿÿÿÿsexÿÿÿÿcount0ÿÿÿÿcount1ÿÿÿÿcount2
ÿÿ1.ÿ0ÿÿÿÿÿÿÿ1119ÿÿÿÿÿ960ÿÿÿÿ1840
ÿÿ2.ÿ1ÿÿÿÿÿÿÿ.ÿÿÿÿÿÿÿ.ÿÿÿÿÿÿÿ.
ÿÿ3.ÿend

.ÿ
.ÿlocalÿtotalÿ=ÿ6069

.ÿquietlyÿreplaceÿcount0ÿ=ÿround(0.318ÿ*ÿ`total')ÿifÿsex

.ÿquietlyÿreplaceÿcount1ÿ=ÿround(0.240ÿ*ÿ`total')ÿifÿsex

.ÿquietlyÿreplaceÿcount2ÿ=ÿround(0.442ÿ*ÿ`total')ÿifÿsex

.ÿassertÿcount0ÿ+ÿcount1ÿ+ÿcount2ÿ==ÿcond(sex,ÿ6069,ÿ3919)

.ÿ
.ÿegenÿintÿtotalÿ=ÿrowtotal(count?)

.ÿrenameÿ(count?)ÿcount?1

.ÿforvaluesÿiÿ=ÿ0/2ÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿgenerateÿintÿcount`i'0ÿ=ÿtotalÿ-ÿcount`i'1
ÿÿ3.ÿ}

.ÿquietlyÿreshapeÿlongÿcount0ÿcount1ÿcount2,ÿi(sex)ÿj(pos)

.ÿquietlyÿreshapeÿlongÿcount,ÿi(sexÿpos)ÿj(sco)

.ÿ
.ÿlabelÿdefineÿSexesÿ0ÿFÿ1ÿM

.ÿlabelÿvaluesÿsexÿSexes

.ÿ
.ÿlabelÿdefineÿHungerÿ0ÿ"Experienceÿhunger"ÿ1ÿ"Atÿriskÿofÿhunger"ÿ2ÿ"Foodÿsecure"

.ÿlabelÿvaluesÿscoÿHunger

.ÿ
.ÿlabelÿdefineÿNYÿ0ÿNÿ1ÿY

.ÿlabelÿvaluesÿposÿNY

.ÿ
.ÿlist,ÿnoobsÿsepby(sex)

ÿÿ+-----------------------------------------------+
ÿÿ|ÿsexÿÿÿposÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿscoÿÿÿcountÿÿÿtotalÿ|
ÿÿ|-----------------------------------------------|
ÿÿ|ÿÿÿFÿÿÿÿÿNÿÿÿExperienceÿhungerÿÿÿÿ2800ÿÿÿÿ3919ÿ|
ÿÿ|ÿÿÿFÿÿÿÿÿNÿÿÿAtÿriskÿofÿhungerÿÿÿÿ2959ÿÿÿÿ3919ÿ|
ÿÿ|ÿÿÿFÿÿÿÿÿNÿÿÿÿÿÿÿÿÿFoodÿsecureÿÿÿÿ2079ÿÿÿÿ3919ÿ|
ÿÿ|ÿÿÿFÿÿÿÿÿYÿÿÿExperienceÿhungerÿÿÿÿ1119ÿÿÿÿ3919ÿ|
ÿÿ|ÿÿÿFÿÿÿÿÿYÿÿÿAtÿriskÿofÿhungerÿÿÿÿÿ960ÿÿÿÿ3919ÿ|
ÿÿ|ÿÿÿFÿÿÿÿÿYÿÿÿÿÿÿÿÿÿFoodÿsecureÿÿÿÿ1840ÿÿÿÿ3919ÿ|
ÿÿ|-----------------------------------------------|
ÿÿ|ÿÿÿMÿÿÿÿÿNÿÿÿExperienceÿhungerÿÿÿÿ4139ÿÿÿÿ6069ÿ|
ÿÿ|ÿÿÿMÿÿÿÿÿNÿÿÿAtÿriskÿofÿhungerÿÿÿÿ4612ÿÿÿÿ6069ÿ|
ÿÿ|ÿÿÿMÿÿÿÿÿNÿÿÿÿÿÿÿÿÿFoodÿsecureÿÿÿÿ3387ÿÿÿÿ6069ÿ|
ÿÿ|ÿÿÿMÿÿÿÿÿYÿÿÿExperienceÿhungerÿÿÿÿ1930ÿÿÿÿ6069ÿ|
ÿÿ|ÿÿÿMÿÿÿÿÿYÿÿÿAtÿriskÿofÿhungerÿÿÿÿ1457ÿÿÿÿ6069ÿ|
ÿÿ|ÿÿÿMÿÿÿÿÿYÿÿÿÿÿÿÿÿÿFoodÿsecureÿÿÿÿ2682ÿÿÿÿ6069ÿ|
ÿÿ+-----------------------------------------------+

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿglmÿposÿi.sco##i.sexÿ[fweight=count],ÿfamily(binomial)ÿlink(identity)ÿnolog

GeneralizedÿlinearÿmodelsÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿ=ÿÿÿÿÿ29,964
Optimizationÿÿÿÿÿ:ÿMLÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿResidualÿdfÿÿÿÿÿ=ÿÿÿÿÿ29,958
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿScaleÿparameterÿ=ÿÿÿÿÿÿÿÿÿÿ1
Devianceÿÿÿÿÿÿÿÿÿ=ÿÿ37081.86667ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ(1/df)ÿDevianceÿ=ÿÿÿ1.237795
Pearsonÿÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿÿÿÿ29964ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ(1/df)ÿPearsonÿÿ=ÿÿÿÿÿ1.0002

Varianceÿfunction:ÿV(u)ÿ=ÿu*(1-u)ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ[Bernoulli]
Linkÿfunctionÿÿÿÿ:ÿg(u)ÿ=ÿuÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ[Identity]

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿAICÿÿÿÿÿÿÿÿÿÿÿÿÿ=ÿÿÿ1.237948
Logÿlikelihoodÿÿÿ=ÿ-18540.93334ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿBICÿÿÿÿÿÿÿÿÿÿÿÿÿ=ÿÿ-271717.8

--------------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿOIM
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿposÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
---------------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿscoÿ|
ÿÿAtÿriskÿofÿhungerÿÿ|ÿÿ-.0405716ÿÿÿ.0099624ÿÿÿÿ-4.07ÿÿÿ0.000ÿÿÿÿ-.0600975ÿÿÿ-.0210456
ÿÿÿÿÿÿÿÿFoodÿsecureÿÿ|ÿÿÿ.1839755ÿÿÿ.0107522ÿÿÿÿ17.11ÿÿÿ0.000ÿÿÿÿÿ.1629016ÿÿÿÿ.2050494
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿsexÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿMÿÿ|ÿÿÿ.0324775ÿÿÿ.0093697ÿÿÿÿÿ3.47ÿÿÿ0.001ÿÿÿÿÿ.0141133ÿÿÿÿ.0508417
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿsco#sexÿ|
Atÿriskÿofÿhunger#Mÿÿ|ÿÿ-.0373655ÿÿÿÿ.012847ÿÿÿÿ-2.91ÿÿÿ0.004ÿÿÿÿ-.0625452ÿÿÿ-.0121858
ÿÿÿÿÿÿFoodÿsecure#Mÿÿ|ÿÿ-.0600671ÿÿÿ.0138558ÿÿÿÿ-4.34ÿÿÿ0.000ÿÿÿÿ-.0872239ÿÿÿ-.0329103
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿÿ.285532ÿÿÿ.0072149ÿÿÿÿ39.58ÿÿÿ0.000ÿÿÿÿÿ.2713911ÿÿÿÿÿ.299673
--------------------------------------------------------------------------------------
Coefficientsÿareÿtheÿriskÿdifferences.

.ÿcontrastÿsco@sex

Contrastsÿofÿmarginalÿlinearÿpredictions

Marginsÿÿÿÿÿÿ:ÿasbalanced

------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿÿchi2ÿÿÿÿÿP>chi2
-------------+----------------------------------
ÿÿÿÿÿsco@sexÿ|
ÿÿÿÿÿÿÿÿÿÿFÿÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿ493.66ÿÿÿÿÿ0.0000
ÿÿÿÿÿÿÿÿÿÿMÿÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿ577.45ÿÿÿÿÿ0.0000
ÿÿÿÿÿÿJointÿÿ|ÿÿÿÿÿÿÿÿÿÿ4ÿÿÿÿÿ1071.12ÿÿÿÿÿ0.0000
------------------------------------------------

.ÿ
.ÿ//ÿÿor
.ÿlogitÿposÿi.sco##i.sexÿ[fweight=count],ÿnolog

LogisticÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿ29,964
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿLRÿchi2(5)ÿÿÿÿÿÿÿÿ=ÿÿÿÿ1063.15
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿchi2ÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0000
Logÿlikelihoodÿ=ÿ-18540.933ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿPseudoÿR2ÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0279

--------------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿposÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
---------------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿscoÿ|
ÿÿAtÿriskÿofÿhungerÿÿ|ÿÿ-.2084894ÿÿÿ.0512876ÿÿÿÿ-4.07ÿÿÿ0.000ÿÿÿÿ-.3090112ÿÿÿ-.1079675
ÿÿÿÿÿÿÿÿFoodÿsecureÿÿ|ÿÿÿ.7950626ÿÿÿ.0476999ÿÿÿÿ16.67ÿÿÿ0.000ÿÿÿÿÿ.7015725ÿÿÿÿ.8885526
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿsexÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿMÿÿ|ÿÿÿ.1542498ÿÿÿÿ.044839ÿÿÿÿÿ3.44ÿÿÿ0.001ÿÿÿÿÿ.0663669ÿÿÿÿ.2421326
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿsco#sexÿ|
Atÿriskÿofÿhunger#Mÿÿ|ÿÿ-.1808585ÿÿÿ.0655235ÿÿÿÿ-2.76ÿÿÿ0.006ÿÿÿÿ-.3092821ÿÿÿ-.0524349
ÿÿÿÿÿÿFoodÿsecure#Mÿÿ|ÿÿ-.2655101ÿÿÿ.0608533ÿÿÿÿ-4.36ÿÿÿ0.000ÿÿÿÿ-.3847803ÿÿÿ-.1462399
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿ-.917184ÿÿÿ.0353666ÿÿÿ-25.93ÿÿÿ0.000ÿÿÿÿ-.9865013ÿÿÿ-.8478667
--------------------------------------------------------------------------------------

.ÿcontrastÿsco@sex

Contrastsÿofÿmarginalÿlinearÿpredictions

Marginsÿÿÿÿÿÿ:ÿasbalanced

------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿÿchi2ÿÿÿÿÿP>chi2
-------------+----------------------------------
ÿÿÿÿÿsco@sexÿ|
ÿÿÿÿÿÿÿÿÿÿFÿÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿ492.33ÿÿÿÿÿ0.0000
ÿÿÿÿÿÿÿÿÿÿMÿÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿ554.24ÿÿÿÿÿ0.0000
ÿÿÿÿÿÿJointÿÿ|ÿÿÿÿÿÿÿÿÿÿ4ÿÿÿÿÿ1046.57ÿÿÿÿÿ0.0000
------------------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

Because the categories are ordered, you can also use something like a Helmert or reverse-Helmert contrast. See the help file for -contrast- for more information.

[Carlo Lazzaro]

Siluleko:
as you surmise, there's no scope in comparing the sum of the the proportions, as they sum up to 100% for both males and females.
You can replicate -prtesti- for the remaining two strata (ie, -At risk of hunger- and -Food insecure-) keeping the same N but changing the proportions as necessary.

[Joseph Coveney]

Replace the value labels for sex above with

label define Sexes 0 M 1 F

I inadvertently flipped the two rows in the post from where I copied the counts and proportions.

Actually, you'll need to give more information about the counts in each cell, too, and take them into consideration in order to have the totals add up across rows. But the modeling is done in the same way.

[Joseph Coveney]

there's no scope in comparing the sum of the the proportions, as they sum up to 100% for both males and females.

As Carlo noticed, and I neglected to, the proportions in the cells are proportions of the total. So disregard what I show above.

[Siluleko Mkhize]

Siluleko:
as you surmise, there's no scope in comparing the sum of the the proportions, as they sum up to 100% for both males and females.
You can replicate -prtesti- for the remaining two strata (ie, -At risk of hunger- and -Food insecure-) keeping the same N but changing the proportions as necessary.

Okay, that will do. I think I should do three comparisons (food secure vs at risk, food secure vs food insecure and finally food insecure vs at risk).
I will then find a way to show the three P values or use symbols to signal significant and non-significant comparisons.
It may seem iterative compared to doing a post hoc for multiple comparisons

[Carlo Lazzaro]

Siluleko:
just echoing Nick Cox 's wise comment in #2, you should decide which of your variables is the outcome or the predictors.
My idea in #7 was that you were planning to compare the prevalence of the three strata of -food security- in males vs females.
In #14 you seem to have a second thought about your research goal, that is making between strata comparisons.
I think you should first clarify what is the aim of your research.