Using the wealth index and wealth quintile variables in DHS dataset

[George Kariuki]

Hallo great friends,

May be many of you here have great knowledge of the Demographic and health survey data. I want to use the wealth index variable to determine the what influences poverty in Kenya. This Variable is released with DHS datasets as of 2003 or 2008 onwards. I intend to get a binary variable "poor==0, rich==1" for use in a logit model. Does the fact that there is always a larger (over 80%) sample of women than men affect any household level analysis?

Thank you.

Georges.

[Sergiy Radyakin]

Keyword is "any". I can imagine an analysis where that would matter. What is that you are doing specifically?

Note that DHS was traditionally conditioned on having at least 1 woman in the household. So men living alone, or men living together were not covered by DHS. Check particular sampling procedure for your country and years.

[Friedrich Huebler]

Does the fact that there is always a larger (over 80%) sample of women than men affect any household level analysis?

You may be thinking of the larger sample size of the Individual Recode file with women's data compared to the smaller sample size of the Male Recode file with men's data. In the Household Member Recode file the distribution by sex is much closer to 50:50. For a description of DHS dataset types see this link.

As an example, below is the unweighted number of male and female observations in the Household Member Recode file of the 2008-09 DHS in Kenya.

. tab hv104, m

     Sex of |
  household |
     member |      Freq.     Percent        Cum.
------------+-----------------------------------
       Male |     18,774       48.74       48.74
     Female |     19,741       51.26      100.00
------------+-----------------------------------
      Total |     38,515      100.00

[George Kariuki]

Thank you Friedrich. You are right, I am tabs on now!
Cheers good friends.
George.

[George Kariuki]

Hey Friedrich,

Do you know what goes into the "wealth index" in the DHS data? Is it equivalent to any form of asset indices? How is it computed?

Thanks for accepting to help.

[Friedrich Huebler]

The page at the following URL is the first result of a Google search for "DHS wealth index": http://www.dhsprogram.com/topics/wealth-index/Index.cfm

[George Kariuki]

You seem to be my savior Friedrich. I keep coming back. The education variable. I have gone through the recode document that defines the variables in the DHS. My household level analysis will I want to use the years of education of the household head as one of the explanatory variables determining whether that household is either poor or not poor. I however cant find a way to get the individual characteristic (education years) and that individual be the HH head. Please help.

[Friedrich Huebler]

In a DHS household member recode file, the household head is usually identified in the variable hv101, "Relationship to head". Household member recode files also usually contain a variable hv108, "Education in single years". You could have found these variables with the lookfor command.

. lookfor head

              storage   display    value
variable name   type    format     label      variable label
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
hv218           byte    %8.0g                 Line number of head of househ.
hv219           byte    %8.0g      HV219      Sex of head of household
hv220           byte    %8.0g      HV220      Age of head of household
hv101           byte    %24.0g     HV101      Relationship to head

. lookfor years

              storage   display    value
variable name   type    format     label      variable label
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
hv108           byte    %12.0g     HV108      Education in single years
hv124           byte    %12.0g     HV124      Education in single years - current school year
hv128           byte    %12.0g     HV128      Education in single years - previous school year
ha1             byte    %8.0g                 Women's age in years
hb1             byte    %8.0g                 Men's age in years

Below are the values of the variables hv101 and hv108 in the household member recode file from the 2008-09 DHs in Kenya.

. tab hv101, m

    Relationship to head |      Freq.     Percent        Cum.
-------------------------+-----------------------------------
                    Head |      9,057       23.52       23.52
         Wife or husband |      4,812       12.49       36.01
            Son/daughter |     17,970       46.66       82.67
     Son/daughter-in-law |        309        0.80       83.47
              Grandchild |      3,038        7.89       91.36
                  Parent |        186        0.48       91.84
           Parent-in-law |         69        0.18       92.02
          Brother/sister |        691        1.79       93.81
          Other relative |        699        1.81       95.63
    Adopted/foster child |        225        0.58       96.21
             Not related |        543        1.41       97.62
   Niece/nephew by blood |        674        1.75       99.37
Niece/nephew by marriage |        232        0.60       99.97
                      DK |          4        0.01       99.98
                      99 |          6        0.02      100.00
-------------------------+-----------------------------------
                   Total |     38,515      100.00

. tab hv108, m

Education in |
single years |      Freq.     Percent        Cum.
-------------+-----------------------------------
           0 |     13,611       35.34       35.34
           1 |      1,584        4.11       39.45
           2 |      1,635        4.25       43.70
           3 |      1,811        4.70       48.40
           4 |      1,964        5.10       53.50
           5 |      1,813        4.71       58.21
           6 |      2,006        5.21       63.41
           7 |      2,937        7.63       71.04
           8 |      3,837        9.96       81.00
           9 |        894        2.32       83.32
          10 |        875        2.27       85.60
          11 |      1,262        3.28       88.87
          12 |      2,229        5.79       94.66
          13 |        407        1.06       95.72
          14 |        562        1.46       97.18
          15 |        293        0.76       97.94
          16 |         46        0.12       98.06
          17 |         88        0.23       98.28
          18 |        190        0.49       98.78
          19 |        234        0.61       99.38
          20 |         74        0.19       99.58
          21 |         35        0.09       99.67
          22 |          8        0.02       99.69
          23 |          6        0.02       99.70
          24 |          3        0.01       99.71
          25 |          1        0.00       99.71
          DK |         25        0.06       99.78
          99 |         85        0.22      100.00
-------------+-----------------------------------
       Total |     38,515      100.00

[George Kariuki]

Thank you so much. I see that it also exists under the household recode file by just adding _01 on the variable

hv108

[George Kariuki]

Friedrich, strange results, I don't know why. I included the age of household head variable and its quardratic in a logit regression. Turns out that the age of HH head is not significant in explaining whether a HH is poor or rich. Further, the turning point is surprisingly at 14. Could I be doing something wrong? the logit command doesn't accept the aweights which I thik should be correcting for sampling disproportions. Check this results

. logit povertydmy hhage hhage2 hhsize educyrs i.gender i.marstat i.hhloc i.region

Iteration 0:   log likelihood = -6201.4769 
Iteration 1:   log likelihood = -3416.3541 
Iteration 2:   log likelihood = -3305.7253 
Iteration 3:   log likelihood = -3296.1457 
Iteration 4:   log likelihood = -3295.1891 
Iteration 5:   log likelihood =  -3295.159 
Iteration 6:   log likelihood =  -3295.159 

Logistic regression                               Number of obs   =       8950
                                                  LR chi2(16)     =    5812.64
                                                  Prob > chi2     =     0.0000
Log likelihood =  -3295.159                       Pseudo R2       =     0.4686

--------------------------------------------------------------------------------------------------
                      povertydmy |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------------------------+----------------------------------------------------------------
                           hhage |  -.0021826    .011318    -0.19   0.847    -.0243654    .0200001
                          hhage2 |   .0000745   .0001125     0.66   0.508     -.000146     .000295
                          hhsize |  -.1645016   .0148169   -11.10   0.000    -.1935421    -.135461
                         educyrs |   .2206107   .0090927    24.26   0.000     .2027893    .2384321
                                 |
                          gender |
                        1. male  |  -.1591968   .0779198    -2.04   0.041    -.3119169   -.0064768
                                 |
                         marstat |
           2. currently married  |  -.3263309   .1417641    -2.30   0.021    -.6041834   -.0484784
                     3. widowed  |  -.6101006   .1748029    -3.49   0.000     -.952708   -.2674932
4. divorced/not living together  |  -.3313861   .1838207    -1.80   0.071     -.691668    .0288959
                                 |
                           hhloc |
                       1. urban  |   3.429048    .107804    31.81   0.000     3.217756     3.64034
                                 |
                          region |
                     2. Central  |  -2.084518   .7209216    -2.89   0.004    -3.497498   -.6715377
                       3. Coast  |   -2.97589   .7196659    -4.14   0.000    -4.386409   -1.565371
                     4. Eastern  |  -2.684644   .7210483    -3.72   0.000    -4.097873   -1.271416
                      5. Nyanza  |  -3.284647   .7203815    -4.56   0.000    -4.696569   -1.872726
                  6. Rift valey  |  -2.872649   .7204791    -3.99   0.000    -4.284762   -1.460536
                     7. Western  |  -3.654501   .7206855    -5.07   0.000    -5.067019   -2.241984
                8. Northeastern  |  -4.103123   .7299678    -5.62   0.000    -5.533833   -2.672412
                                 |
                           _cons |   1.612365   .7619821     2.12   0.034     .1189074    3.105822
--------------------------------------------------------------------------------------------------

[George Kariuki]

Computing the turning point on age while looking at how age dynamically predicts probabilities for household poverty we take the beta of the age variable, negate it and divide by twice the quadratic beta. Simply the first order condition. Check the outcome:

. di -(-.0021826)/(2*.0000745)
14.648322

Past literature shows pretty more sensible turning points which makes me feel there is something sinister with my working. Thank you in advance for accepting to give ideas.

[Friedrich Huebler]

Please start a new thread with the questions in posts #10 and #11. The likelihood of an answer is also increased if you don't address your questions to one individual on Statalist.

[George Kariuki]

Thank you. I thought of writing to you specifically since this seems more of an analytical question than a Stata use command. From your earlier help I deemed it possible to have an incite fro you but yeah I get the gist.. let me create a new thread for this.