# How can I accurately calculate the median family income by state for families below 200% of the poverty level?

Hello Michelle,

I would be supremely grateful if you could help me out.

I’ve downloaded a dataset that includes the following variables and am using SPSS:

YEAR

DATANUM

SERIAL

PERNUM

HHINCOME

INCTOT

FTOTINC

POVERTY

AGE

HHWT

HHTYPE

STATEFIP

GQ

PERWT

I attempted to select cases where PERNUM was = 1 and POVERTY <= 200 and NE 0. I used the PERWT. Are there other variables I need to manipulate, like GQ or HHTYPE?

Since ,I wasn’t sure if my numbers were correct, I tried to replicate basic median family income by state and was unable to replicate the numbers here: https://factfinder.census.gov/faces/t…

I know the IPUMS data aren’t supposed to match the FactFinder data exactly, but my figures seemed a bit off: They are below:

Alabama 52000

Arizona 55000

Arkansas 48600

California 68000

Connecticut 82800

Delaware 70000

District of Columbia 70000

Florida 53000

Georgia 57000

Hawaii 84000

Idaho 58000

Illinois 68000

Indiana 58000

Iowa 63100

Kansas 61000

Kentucky 50000

Louisiana 50000

Maine 57600

Maryland 84000

Massachusetts 83400

Michigan 58500

Minnesota 72000

Mississippi 46700

Missouri 57000

Montana 56300

New Hampshire 80000

New Jersey 85000

New Mexico 49200

New York 68000

North Carolina 54000

North Dakota 68000

Ohio 58100

Oklahoma 53000

Oregon 60000

Pennsylvania 62960

Rhode Island 66200

South Carolina 52300

South Dakota 60200

Tennessee 53000

Texas 60000

Utah 70200

Vermont 65000

Virginia 73500

Washington 69900

West Virginia 48800

Wisconsin 64000

Wyoming 64020

There are a couple important details that you should keep in mind while performing this analysis. First, the POVERTY variable does exclude group quarters. So, you’ll want to drop observations with GQ==3 through 6. Second, PERNUM identifies the person number within the household, rather than family. This is an important detail because the POVERTY variable uses family income, rather than household income. So by limiting your data to only cases with PERNUM==1 you are actually dropping any families that are denoted as the “second family” within a given household. This is likely one reason why your estimates are slightly different than those found via American Fact Finder. The family-level equivalent to PERNUM is the SFRELATE variable. In this variable the U.S. Census defines one person per sub-family to be a “reference person”, this is a similar concept to “household head”. So, if you rerun your analysis, include SFRELATE and filter for SFRELATE==1. This should get you closer to the estimates in American Fact Finder. Third, in order to obtain accurate representative statistical estimates at the sub-family level you’ll need to use the person level sample weight PERWT.