The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Malaria » How to make and interpreting Bivariate statistics analysis?
How to make and interpreting Bivariate statistics analysis? [message #11402] Mon, 12 December 2016 14:14 Go to next message
hamzah is currently offline  hamzah
Messages: 2
Registered: November 2016
Location: Indonesia
Member

Dear experts

Regarding statistics to population survey, could you please tell me which one of the syntax using for bivariate analysis [chi square] and what does different the meaning of each syntax like below:

1.
svy: tabulate sex malaria
and output here : 
Number of strata   =         1                 Number of obs     =     259,885
Number of PSUs     =     4,418                 Population size   =  30,152,652
                                               Design df         =       4,417

-------------------------------
gender of |
responden |       malaria      
ts        |    no    yes  Total
----------+--------------------
     male | .4744  .0185  .4929
   female | .4909  .0162  .5071
          | 
    Total | .9653  .0347      1
-------------------------------
  Key:  cell proportion

  Pearson:
    Uncorrected   chi2(1)         =   58.3020
    Design-based  F(1, 4417)      =   49.6352     P = 0.0000


2.
.  svy: tabulate sex malaria, row
and output here : 
 (running tabulate on estimation sample)

Number of strata   =         1                 Number of obs     =     259,885
Number of PSUs     =     4,418                 Population size   =  30,152,652
                                               Design df         =       4,417

-------------------------------
gender of |
responden |       malaria      
ts        |    no    yes  Total
----------+--------------------
     male | .9625  .0375      1
   female |  .968   .032      1
          | 
    Total | .9653  .0347      1
-------------------------------
  Key:  row proportion

  Pearson:
    Uncorrected   chi2(1)         =   58.3020
    Design-based  F(1, 4417)      =   49.6352     P = 0.0000




3.
. svy linearized : tabulate sex  malaria, obs row percent ci

and output here : 
 (running tabulate on estimation sample)

Number of strata   =         1                 Number of obs     =     259,885
Number of PSUs     =     4,418                 Population size   =  30,152,652
                                               Design df         =       4,417

-------------------------------------------------------
gender of |
responden |                   malaria                  
ts        |            no            yes          Total
----------+--------------------------------------------
     male |         96.25          3.746            100
          | [96.01,96.48]  [3.518,3.987]               
          |       1.2e+05           5595        1.3e+05
          | 
   female |          96.8          3.198            100
          | [96.57,97.02]  [2.979,3.431]               
          |       1.3e+05           4971        1.3e+05
          | 
    Total |         96.53          3.468            100
          | [96.31,96.74]  [3.257,3.692]               
          |       2.5e+05        1.1e+04        2.6e+05
-------------------------------------------------------
  Key:  row percentage
        [95% confidence interval for row percentage]
        number of observations

  Pearson:
    Uncorrected   chi2(1)         =   58.3020
    Design-based  F(1, 4417)      =   49.6352     P = 0.0000



How to make odds ratio for cross-sectional design survey? Should I make syntax for prevalence ratio or may I take directly odds ratio in the syntax below?

5.
. svy linearized : logistic sex malaria

and output here : 
 (running logistic on estimation sample)

Survey: Logistic regression

Number of strata   =         1                 Number of obs     =     259,885
Number of PSUs     =     4,418                 Population size   =  30,152,652
                                               Design df         =       4,417
                                               F(   1,   4417)   =       49.54
                                               Prob > F          =      0.0000

------------------------------------------------------------------------------
             |             Linearized
         sex | Odds Ratio   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     malaria |   .8488294   .0197667    -7.04   0.000     .8109481    .8884803
       _cons |   1.034818   .0042681     8.30   0.000     1.026484    1.043219
------------------------------------------------------------------------------



Based on the table above [chi square and binary logistic].

Where the sex variable which assumptions male is given code = 0 and female is given code = 1.
Malaria prevalence differs by sex Males are more likely to have malaria than females (1.85% males versus 1.62% females, P = 0.000). Based on odds ratio (OR) female have the chances of getting malaria 0.85% or 0.85 times than male (as categorical reference)

How do I interpret an odds ratio less than 1 in a logistic regression?
May I will be written male with a chance of 1 / 0.85 times or 1.2 times to get malaria compared than female as well?

or

The odds of malaria in male decreased by (1 - 0.85 ) 15% compared those in a female. Whatever on the dependent variable decreases. For each unit increase, it decreases by a multiple of (1 - OR )


Thank you in advance for your reply




Sincerely yours,


Hamzah

Re: How to make and interpreting Bivariate statistics analysis? [message #11459 is a reply to message #11402] Fri, 23 December 2016 15:43 Go to previous message
Liz-DHS
Messages: 1516
Registered: February 2013
Senior Member
Dear User,

We recommend that you search online for some resources to better understand the output you are generating. This site is a good reference with lots of examples and help with interpretation:
http://statistics.ats.ucla.edu/stat/stata/

A few notes on your output:
1) Without more information on the dataset you are using and on your analytic objectives it is difficult to provide much guidance but it appears that your sample size is unusually large. Is this a pooled sample from multiple surveys?
2) It also appears that you have misrepresented the sample design in your survey set command. It would be very unusual for a DHS or MIS survey to have only one stratum.
3) I would also suggest verifying that you are properly specifying the denominator for this analysis. In most DHS/MIS surveys, only children 6-59 months of age who spent the previous night in interviewed households (de facto) were eligible for malaria parasitemia testing.
4) Please check that you have properly recoded the sex variable. In our standard recode files 1=male and 2=female.
5) For proper interpretation of ORs <1.0 please see online resources such as the one given above.

Good luck!

Previous Topic: Reproduce household ownership of ITN's using household member recode files
Next Topic: Mosquito nets
Goto Forum:
  


Current Time: Thu Mar 28 18:18:47 Coordinated Universal Time 2024