A to Z
Data &
Forms &
News &
Licensing &
Rules &
Public Health
Oregon PRAMS Report; 1998-99

Oregon PRAMS: First Year Report, 1998-99

Appendix C:

Weighting Methodology

Note that the data presented here are weighted data: estimates of how a random sample of Oregon women would answer these questions. The data represent information from women who answered the questions in all six strata, and are weighted for state-wide averages. It will be possible to analyze the answers to PRAMS questions for any subset of sampled women, including by birth weight or race/ethnicity (the oversampled strata).

I. Sampling

The 1997 total births from Oregon resident mothers who gave birth in Oregon was used as the basis for determining sample sizes. This population was broken down into normal birthweight (NBW, birthweight > 2500g) and low birthweight (LBW, birthweight < 2500g); and race/ethnicity categories. The Oregon birth certificate race/ethnicity categories are different from CDC's categories (White, Hispanic, Black, American Indian, and Asian/Pacific Islander) and program codes were created to match Oregon codes to CDC-PRAMS categories.

CDC's proposed initial allocation was 400 for each race/ethnicity category per year. Assignment to normal and low birthweight in each race/ethnicity group was based on the derived population proportions. Oregon PRAMS oversampled White Low Birthweight to match White Normal Birthweight at 400 each. Hence, the total initial allocation for the state was 2,400 per year. The finite population correction factor was used to adjust allocations where the sampling fraction exceeded 1 in 10. Hence, all groups except NBW White and Hispanic were adjusted using the finite correction factor. After this, the resulting group sample size was inflated by dividing by 70%, the intended response rate. The group sample sizes were transformed into the resulting fractions (expressed as decimals) for each race/ethnicity category and ultimately expressed as common, easy to remember fractions. These fractions were multiplied with the actual counts for each of the 10 birthweight and race/ethnicity groups to determine the actual estimated yearly sample size. Dividing the result by 12 gives an estimate for the monthly sample size. The actual procedure was to follow these series of steps for each month.

Sampling Proportions:

  • NBW White: 1/65
  • LBW White: 1/4
  • Hispanic: 1/10
  • African Am: 9/20
  • Am Indian/Alaskan: 1/2
  • Asian/Pacific Islander: 1/4
Most of the mothers sampled in each month had given birth 60 to 90 days prior to the drawing date. However, mothers who were not included in the sampling frame but were eligible in past drawings were given a chance to be picked in succeeding drawings so long as they delivered their babies no more than 180 days ago. These occur primarily when there are late birth certificate entries at Vital Statistics. Multiple births were noted and the mother was given only one chance to be in the sampling frame. If picked, the Oregon PRAMS cover letter asks the mother to answer the questions as they pertain to one specific baby. Live births which resulted in early death for the babies were also noted. A different cover letter was sent, but the same questionnaire was provided.

In the first year (November 1998 to October 1999) we mailed the survey to 2,919 women. This number was bigger than the initial projections and may be primarily due to the increase in the number of total births and the policy of rounding up to the next higher integer in each of the 10 sampling categories. A total of 1,867 (64.0%)responses were received either in the first or second mailing, or in the telephone interviews.

II. Weighting

Weighting is done in three steps. Weight I represents the over-sampling and strata adjustment, Weight II the unit non-response adjustment, and Weight III the non-coverage adjustment. The final weight is the product of these three weights. The final report represents the weighted percentages of the different responses to each question.

A. Weight I: Over-sampling and strata adjustment

The stratification weight is the total number of eligible mothers for the year divided by the actual number sampled.

B. Weight II: Unit Non-response

CDC identified the variables marital status, high education, low education, parity, high age, low age, and 1st trimester prenatal care as possible indicators of a mother's inclination to respond. Each one of these variables were defined in the following manner:

MARSTAT 0 All Else Marital Status
1 Married
HIGHEDUC 0 At most K12 High Education
1 College +
LOWEDUC 0 At least K12 Low Education
1 < K12
PARITY 0 First born Parity
1 Not first born
AGEHIGH 0 All Else High Age
1 30 <= Age < = 49
AGELOW 0 All Else Low Age
1 11 <= Age < = 19
PNCHIGH 0 Late/No PNC Prenatal care in 1st trimester
1 1st Trimester PNC
Another newly defined variable is:
RESPONSE 0 Non-Response
1 Response
The entire sample for the year is used for the binary logistic regression as the first step of Classification and Regression Tree (CART) analysis. The dependent variable is RESPONSE and the independent variables were all the preceding defined variables. The object is to determine groups of mothers with similar demographic characteristics whose response patterns are statistically significantly different from other groups. The CART process is iterative and is continued until there are either no more statistically significantly different partitions based on the 7 variables or when any of the nth partitions or nodes have total responses less than 25. Nodes with similar response patterns are merged and re-checked for any possible partitions.

This procedure is undertaken in each of the 6 strata (NBW White, LBW White, Hispanic, African American, American Indian/Alaskan Native, and Asian and Pacific Islander).

The CART analysis yielded the following partitions:

  1. NBW White
    1. HIGHEDUC = 1
    2. HIGHEDUC = 0 and AGEHIGH = 0
    3. HIGHEDUC = 0 and AGEHIGH = 1
  2. LBW White
    1. MARSTAT = 0
    2. MARSTAT = 1
  3. Hispanic
    1. MARSTAT = 0
    2. MARSTAT = 1
  4. African American
    1. LOWEDUC = 0
    2. LOWEDUC = 1
  5. American Indian/Alaskan Native
    1. PNCHIGH = 0
    2. PNCHIGH = 1 and (MARSTAT = 1 or (MARSTAT = 0 and AGELOW = 1))
    3. PNCHIGH = 1 and MARSTAT = 0 and AGELOW = 0
  6. Asian/Pacific Islander
    1. LOWEDUC = 0
    2. LOWEDUC = 1
The corresponding weights for each one of the partitions is the total partition or node sample size divided by the number of respondents.

C. Weight III: Non-coverage adjustment

Weight III is the ratio of the current birth certificate file list of all eligible births during the first year to the cumulative sampling frame in the same time period. Ineligible births are those siblings of multiple births which were not chosen to be in the sampling frame.

Return to the top of the page.