Date: Mar 5, 2013 2:24 PM
Author: Richard Ulrich
Subject: Re: application of weights
On Mon, 4 Mar 2013 22:24:19 -0800 (PST), Usman Khan

<usman.cs@gmail.com> wrote:

>Question:

>How should I apply the weights where I have a survey of households and

>

>- I am picking one child out of a household that has both parents living together...

>

>My data looks like this

>

>householdID relationship_to_head age gender

>1 Self 50 M

>1 Wife 49 F

>1 son 10 M

>1 son 8 M

>1 daughter 2 F

>

>2 Self 50 M

>2 Wife 49 F

>2 daughter 12 F

>

>3 Self 50 M

>3 Wife 49 F

>3 son 15 M

>

>4 Self 50 M

>4 Wife 49 F

>

>

>Household Weights are

>1 5

>2 10

>3 10

>4 5

>

>

>now if I were to look at the % Children in households with two parents, would it be

>

> hh_1, hh_2, hh_3 / hh_1, hh_2,hh_3,hh_4 = 5+10+10 / 5+10+10+5

>

>The problem with the above solution is that the weights are for households and not for individual members. Is it fair to use weights in this case?

>

I don't know where you get your weights or why you want to use them.

The only problem that I see with your formula, assuming you like

the weights, is that you have assigned a label that does not

parse properly in English.

The fraction that you compute should be called something like,

"Percent of households that have (one or more) children."

If you want to say something about "individual members" rather

than households, ... Are you referring to only children as

"members"? So, do you want to compute what your label

says, "% children in households with two parents" -- as a

contrast to 0 or 1 parent?

If so -- Your numerator will be a (weighted?) count of children,

and the denominator denominator will be a (weighed?) count

of households that do have one or more child. In your

example, that sum would omit household 4.

--

Rich Ulrich