PRAMS Home | Interactive Data | Definitions | FAQ | Links | CDC PRAMS
SC PRAMS Background and Project Description
Detailed PRAMS Methodology
SC PRAMS Technical Notes
Purpose and History of PRAMS
Since 1987, the Pregnancy Risk Assessment Monitoring System (PRAMS) has served as a state-specific data source for maternal and child health (MCH) issues. PRAMS is a population-based survey of women delivering a live-born infant. This survey collects information on women's experiences and behaviors before, during, and shortly after pregnancy. Thus, states participating in PRAMS gain unique and invaluable information for public health administrators, policy makers, and researchers as they develop programs and policies to improve the health of women and children.
The South Carolina Pregnancy Risk Assessment Monitoring System (SC PRAMS) Project plays a significant role in SC DHEC's public health surveillance activities. The SC PRAMS Project monitors and disseminates information on maternal behavioral risk factors occurring during pregnancy and on a child's early infancy period related to birth outcomes. Thus, the SC PRAMS Project provides sound and reliable maternal and infant health data, which can be used by health professionals for the planning and evaluation of perinatal health programs and for making policy decisions affecting the health of mothers and babies in South Carolina.
The quantitative and qualitative collection, analysis, and use of maternal and child health data are fundamental to the development of an infrastructure to solve women and children's health problems at the state and local levels. Data analysis should be a central component of efforts to identify maternal and child health needs, to design appropriate program interventions, to manage and evaluate those interventions, and to monitor our progress toward achieving selected Healthy People 2010 Objectives. PRAMS is designed not only to generate state-specific data but also to allow comparisons among states through the use of standardized data collection methods.
It is important to remember that information in this module is representative of all South Carolina mothers delivering live infants in South Carolina. Thus, generalizations can be made to this group only. Also, keep in mind that all survey information is based on self-reports from the women.
Purpose of the Pregnancy Risk Assessment Monitoring System (PRAMS)
PRAMS is part of the Centers for Disease Control and Prevention (CDC) initiative to reduce infant mortality and low birthweight. PRAMS is an ongoing, population-based risk factor surveillance system designed to identify and monitor selected maternal experiences and behaviors that occur before and during pregnancy and the child's early infancy among a stratified sample of mothers delivering a live birth.
Epidemiologic surveillance is the ongoing and systematic collection, analysis, and interpretation of health data used for describing and monitoring a health event or behaviors associated with a health event or condition. This information is used for planning, implementing, and monitoring health programs and for informing policy. The dissemination of PRAMS data is an essential step in translating findings from PRAMS into public health action.
PRAMS was developed in 1987 in response to distressing statistics. For example, the U.S. infant mortality rate was no longer declining as rapidly as it had in past years, and the prevalence of low birth weight infants showed little change. At the same time, maternal behaviors such as smoking, drug use, and limited use of prenatal and pediatric care services were recognized as contributors to these slow rates of decline.
PRAMS was initiated to help state health departments establish and maintain an epidemiologic surveillance system of selected maternal behaviors and experiences. PRAMS was designed to supplement data from vital records and to generate data for planning and assessing perinatal health programs in each participating state. Findings from PRAMS are meant to be used to enhance our understanding of maternal behaviors and their relationship with adverse pregnancy outcomes. PRAMS data can also be used to aid in the development and assessment of programs designed to identify high-risk pregnancy and reduce adverse pregnancy outcomes and to inform policy in each participating state.
History of PRAMS
Funding for PRAMS became available to the Division of Reproductive Health, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention in 1987. Funding was made available through cooperative agreements and all state and territorial health departments and the District of Columbia were eligible to apply. In the late summer of 1987, funds were awarded to the District of Columbia, Indiana, Maine, Michigan, Oklahoma, and West Virginia to establish PRAMS surveillance. The PRAMS core questionnaire was developed in 1987 with the participation of numerous individuals within and outside of CDC. To create the questionnaire, potential topics and variables were identified and researched by staff in the Division of Reproductive Health at CDC. Data collection was initiated in the fall of 1988. Because surveillance methods were not well established for the target population of PRAMS, states were encouraged to experiment with different methodologies. For this reason, the first period of data collection from fall 1988 to summer 1990 was considered somewhat of a pilot phase and was referred to as Phase I. South Carolina was not a participant in Phase I.
After a year of data collection, the questionnaire was evaluated and with input from all participating states, it was revised and became Phase II. At this time, the PRAMS software was also updated so that there was one generic version of the software that all states used. In summer 1990, the new software was installed, and shortly thereafter states began using the revised questionnaire. Meanwhile, the state of Alaska had approached CDC about establishing a PRAMS project with their own funding. In the fall of 1990, Alaska became the seventh state to actively collect PRAMS data.
In 1991, additional funds became available through the Infant Health Initiative to expand PRAMS. All of the original states were awarded PRAMS funds, as were six new states: Alabama, Florida, Georgia, New York, South Carolina, and Washington. Alaska was also awarded funds to continue the PRAMS surveillance initiated with state funds. Meanwhile, at the beginning of 1991, California approached CDC about establishing PRAMS surveillance in certain regions of the state using their own funding. Between January and September 1993, the six new states and California began PRAMS data collection. Alabama, Georgia, New York and California included a hospital-based component.
This second funding cycle for PRAMS was established for five years. As operations became more standardized, the concept of a standard protocol was developed. In the fall of 1994, the states convened to determine topic priorities for the questionnaire revision (the Phase 3 questionnaire). For the next year, question modules were developed and tested. PRAMS states implemented the Phase III questionnaire between November 1995 and July 1996. The PRAMS software was upgraded in 1995 to operate in a Windows environment and be LAN compatible. The new software was installed in the fall of 1995.
In fall 1996, funds were awarded for a new five-year PRAMS funding cycle. Although additional funds were not available to expand the number of states, CDC decided to reduce the awards to existing PRAMS states so as to expand PRAMS into new states. Hospital-based surveillance (AL, GA, NY & CA) required more resources per sampled woman than mail/telephone, and therefore, support for the methodology was discontinued in 1996. In addition, two existing PRAMS states were phased out while one state decided not to reapply for funding for the new project period. In October 1996, funds were awarded to ten existing states to continue PRAMS activities and to the five new states of Arkansas, Colorado, Illinois, New Mexico, and North Carolina to establish PRAMS surveillance.
After the 1996 awards, the demand for PRAMS continued to be high. In response to numerous requests for unfunded technical assistance (TA) from states that were interested in proceeding with PRAMS without CDC funding, CDC brought Louisiana into PRAMS as an unfunded TA state. CDC provided all the technical assistance needed to implement the project and Louisiana provided the funding. In 1999, two additional states started PRAMS under this unfunded TA mechanism: Ohio and Utah.
In 1999, additional funds became available for the expansion of PRAMS into new states. Awards were made for a two year funding cycle to five new states: Hawaii, Maryland, Nebraska, New York City, and Vermont. In 2000, funding became available to fund two additional states from the 1999 announcement. Delaware and Mississippi were brought on at that time. CDC also provided federal funding for the three states that were conducting PRAMS under the unfunded technical assistance mechanism: Louisiana, Ohio, and Utah. Revision of the Phase III survey began in 1999 and PRAMS states implemented the Phase IV questionnaire with January 2000 births.
In the fall of 2000, CDC announced expansion of funding for PRAMS under the Safe Motherhood legislation. The value of PRAMS data has been well documented and the demand for PRAMS continues to grow. With this funding cycle, CDC had three aims. First, CDC wanted to expand into new states. Funds were awarded to 31 states to conduct the standard mail/telephone surveillance: six new states (Michigan, Minnesota, New Jersey, Oregon, Rhode Island, and Texas) as well as the 25 states and cities that were already conducting PRAMS. Second, CDC wanted to allow states an opportunity to develop and implement enhanced surveillance activities. Colorado was awarded funds to develop a component to enhance their standard surveillance activities. Third, CDC developed a point-in-time survey to provide an opportunity for states not suited for ongoing surveillance to collect PRAMS data. Montana and North Dakota were funded to conduct point-in-time surveys. PRAMS surveillance currently covers 62% of all U.S. births.
During this funding cycle, one state ceased operations, Colorado completed its enhanced activities in 2003, and Montana and North Dakota completed their point-in-time projects.
With the same process utilized in previous questionnaire evaluations and revisions, the Phase 4 questionnaire was developed and placed in the field in spring 2000. The Phase 4 questionnaire was arranged in two-column format, with instructions to help respondents move through the questionnaire appropriately. Also, the core and standard questions were integrated. Where appropriate, standard questions that relate to core topics appeared with the core questions. Standard questions on topics not covered in the core, as well as all state-developed questions, remained in a separate section at the end of the survey. Following the major revision, the states were provided an opportunity to revise their state-specific questions for 2002 births. About one-half of the states participated in this optional, mini-revision, and their revised Phase 4 questionnaires went into the field in spring 2002.
The Phase 4 questionnaire was evaluated and revised in 2003 with the Phase 5 questionnaire implemented in the states with 2004 births. The core questionnaire was revised slightly, and several new standard questions were developed for states to choose according to their needs. The layout of the questionnaire remained the same, and states still had the option of inserting specific standards questions within the core section. The Phase 5 questionnaire was evaluated in 2007. The Phase 6 questionnaire was revised in 2008 and scheduled for implementation with 2009 births.
In 2006 CDC again received funding to expand PRAMS. Funds were awarded to nine new states as well as the 30 states and cities that were already conducting PRAMS. Eight of the new states (Delaware, Massachusetts, Missouri, Pennsylvania, Tennessee, Virginia, Wisconsin, and Wyoming) were funded to conduct ongoing PRAMS surveillance. South Dakota was funded for a point-in-time survey focusing only on their American Indian population. States participating in PRAMS now account for 75% of all U.S. births.
PRAMS data collection is primarily conducted by mail with telephone follow-up to nonresponders. Prior to 2006, some states developed their own Computer Assisted Telephone Interviewing (CATI) systems to assist in collecting telephone interviews. Others recorded interviews on paper and later keyed them into data entry software. The dual modes used and the variations in CATI systems developed by the states created data management problems for PRAMS.
In 2004, CDC PRAMS contracted with Research Triangle Institute (RTI) for the development and support of a standard CATI system. This change in data collection methodology standardized the way telephone interview data are collected across PRAMS states. The Web-based CATI system takes advantage of automated technology currently available. It collects and generates data files in a consistent manner to facilitate data cleaning and preparation of analysis datasets. All states implemented the standardized CATI during 2006 and 2007. Full implementation of CATI has allowed CDC to provide data to states in a more timely manner, thereby improving the usefulness and effectiveness of PRAMS. Having a more timely and effective surveillance system ultimately supports CDC’s mission of promoting healthier pregnancies and reducing poor birth outcomes.
SC PRAMS Background & Project Description
In 2006, South Carolina’s infant mortality rate was 8.4 deaths per 1,000 live births. From 1989 to 2006, the overall infant mortality rate declined from 12.8 to 8.4. The race specific infant mortality rate for white babies was 5.7 deaths per 1,000 live births in 2006, while for black infants the rate was 13.2 deaths per 1,000 births in 2006. As compared to infants of white mothers, infants of minority mothers are more than twice as likely to die before they reach one year of age. Birthweight is a major determinant of infant death. Infants with a birthweight of less than 2,500 grams (LBW) are at increased risk of death and future chronic disabilities. A comprehensive report on the prevention of low birthweight calls for a better understanding of the behavioral, social, and health service utilization factors that may contribute to the health disparities among minority women and women of lower socioeconomic status.
The SC PRAMS Project, conducted by the Office of Public Health Statistics and Information Services, Surveillance Unit, was established in 1991 through a collaborative agreement between the Centers for Disease Control and Prevention (CDC) and the South Carolina Department of Health and Environmental Control (SC DHEC). The SC PRAMS Project was designed to collect, monitor, analyze, and disseminate information on a wide variety of maternal behaviors and health experiences that may be associated with different birth outcomes.
Approximately 2,300 South Carolina mothers delivering live infants in the state are sampled from the birth registry and surveyed each year. Self-reported information is collected from mail and telephone surveys. The questionnaire consists of 77 structured and standardized questions and is designed to collect information on selected maternal behaviors and experiences during pregnancy and during the child’s early infancy period. The mail survey is sent to sampled mothers up to three times and phone follow-ups are attempted for non-respondents. Sample data is weighted to adjust for sampling design, non-response and non-coverage (see Detailed PRAMS Methodology below). A software package for the statistical analysis of correlated data, SUDAAN, is used to conduct analyses on each year of completed survey data.
Ongoing survey data collection was initiated in January, 1993. Fourteen years of survey data have been completed. The response rates have fluctuated between a low of 66% to a high of 75%, with an overall response rate for all years, 1993-2007, of 70.5%. The response rates for 2006 and 2007 were 66.9% and 68.0%, respectively. The CDC regards a response rate of 70% or above as epidemiologically valid. Since the 2006 and 2007 South Carolina PRAMS response rate is below the 70% threshold, any results obtained using 2006 or 2007 SC PRAMS data should be interpreted with caution.
It is important to remember that information in SC PRAMS reports and modules are representative of all South Carolina mothers delivering live infants in South Carolina. Thus, generalizations can be made to this group only. Also, keep in mind that all survey information is based on self-reports from the women.
Detailed PRAMS Methodology
PRAMS Data Collection Methodology,
One of the strengths of PRAMS is that each participating state uses a standardized data collection methodology. This standardized approach allows for comparisons among states and for optimal use of the data for single-state or multi state analysis. Each state follows the standardized data collection methodology but can also customize some portions of it to tailor the procedures to the needs of the state. For example, the basic methodology recommends three mailings of the questionnaire packet; however, only two mailings are required. PRAMS is a mixed-mode surveillance system that combines two modes of data collection. Each month, a stratified sample is drawn from the current birth certificate file.
For each of these monthly samples, or "batches," a sequence of mail and telephone contacts is attempted. CDC developed and installed a customized tracking system, PRAMTrac, in each state to assist with scheduling mailings and telephone calls, preparing letters, and tracking responses. The primary data collection method is a mail survey. Up to three self-administered surveys are mailed to sampled women, and then mail nonrespondents are contacted by telephone. Here is the sequence of contacts for PRAMS surveillance:
1. Preletter. This letter introduces PRAMS to the sampled mother and informs her that
a questionnaire will soon arrive.
2. Initial Mail Questionnaire Packet. This packet is sent to all sampled mothers 3 to 7 days after the preletter. Its contents are described on the next page.
3. Tickler. The tickler serves as a thank you/reminder note. It is sent 7 to 10 days after the initial mail packet.
4. Second Mail Questionnaire Packet. This packet is sent 7 to 14 days after the tickler to all sampled mothers who did not respond.
5. Third Mail Questionnaire Packet (Optional). This third packet is sent to all remaining nonrespondents 7 to 14 days after the second questionnaire was sent.
6. Telephone Follow-Up. Telephone follow up is initiated for all nonrespondents 7 to 14 days after the mailing of the last questionnaire.
The series of mailings commences 2 to 6 months after the mother delivers her infant. Most mothers respond within 3 to 5 months after giving birth. The questionnaire contains items asking about the early postpartum period; thus, the mailings are timed to ensure that all women can respond for this period. The data collection cycle from the mailing of the preletter to the close of telephone follow up lasts approximately 65 to 85 days. The mail questionnaire packet contains several items.
* A multipurpose cover letter describes PRAMS, explains how and why the mother was chosen, elicits the mother's cooperation, describes procedures for filling out and returning the questionnaire, explains any incentive or reward, and provides a telephone number for additional information. This letter is modified slightly for the second and third mailings, primarily by adding an additional appeal for response.
* Each state's questionnaire booklet is 14 pages long, has a colorful cover designed by the state, is slightly smaller than an 8-1/2" x 11" sheet of paper, and contains an extra page for comments from the mother. A self-addressed return envelope with postage is provided.
* A question-and-answer brochure contains additional information and answers the most frequently asked questions about PRAMS. It can be an important tool to convince the mother to participate.
* A calendar serves as a memory aid for answering the questions.
* Some type of participation incentive (sent to all sampled mothers) or reward (sent to all respondents) is included in the packet or explained in the cover letter. Examples are coupons for certified birth certificates, participation in a raffle for a cash award, prepaid telephone cards, bibs, cash (a dollar bill), and magnetic picture frames. South Carolina uses a CD incentive and cash reward. Telephone follow-up begins after the last questionnaire is mailed. States use various sources of telephone numbers to obtain valid numbers. Calls to a particular number are staggered over times of the day and days of the week. The calling period for a batch runs from 2 to 3 weeks. Up to 15 attempts are made to contact a mother. Often, telephone interviewers arrange call-back interviews to accommodate the mother's schedule.
The PRAMS Questionnaire
With the help of numerous persons within and outside CDC, the original PRAMS questionnaire was developed in 1987. An extensive list of topics was identified and researched for the questionnaire. From this list, questions were developed and tested and ultimately placed on the questionnaire. Participating states used this questionnaire (Phase 1) from fall 1988 through 1989. After an evaluation of the Phase 1 questionnaire, CDC and the participating PRAMS states developed the Phase 2 questionnaire and put it in the field during 1990. Although the questionnaire maintained its original structure, selected questions were revised, some were deleted, and new questions were added. South Carolina began data collection using Phase II.
In 1994, CDC again collaborated with the participating PRAMS states to develop a Phase 3 questionnaire. This revision was more extensive than the revision of 1989. As with the development of the first and second questionnaires, a list of potential topics was extensively researched. Based on this research, questions were revised or dropped, and new questions were added. As before, the questionnaire retained its original structure. In fall 1995 and early 1996, states began to use the Phase 3 questionnaire. States implemented the fourth revision of the questionnaire with January 2000 births. This revision process was repeated before the implementation of the Phase 5 questionnaire in 2004 and again before the implementation of the Phase 6 questionnaire in 2009.
The questionnaire consists of two parts, a core portion that is the same for all states and a state-specific portion tailored to each state's needs. Topics addressed in the PRAMS core questionnaire include barriers to and content of prenatal care, obstetric history, maternal use of alcohol and cigarettes, nutrition, economic status, maternal stress, and early infant development and health status. For the state-specific portion of the questionnaire, states have two options: They can develop and test their own questions, or they can select from a series of "standard" questions on topics that have already been developed and tested by CDC. These questions reflect additional topics of interest to states. They were originally developed during the revision process for Phase 3, and additional questions have been added since that time.
In addition to the questionnaire created for the mail packet, a telephone version of the core and state-specific questions was also developed for telephone interviews. The interviewer-administered questionnaire must be formatted differently than the self-administered questionnaire. It includes prompts and instructions for the interviewer that are not read aloud to the respondent. The interviewer-administered questionnaire format ensures that all interviewers deliver questions and instructions uniformly and consistently with the mail questionnaire. As South Carolina's Hispanic population increases, we have worked with the CDC team to develop Spanish materials for mail and telephone contacts.
PRAMS Weighting Process
Each participating state draws a stratified systematic sample of 100 to 250 new mothers every month from a frame of eligible birth certificates. Most states, including South Carolina, over sample low weight births, and many states stratify by mother's race or ethnicity. South Carolina stratifies by infant birthweight. Annual sample sizes range from 1,500 to 3,200, divided among two to six strata. Mothers' responses are linked to extracted birth certificate data for analysis. Thus, the PRAMS data set also contains a wealth of demographic and medical information collected through the state's vital records system.
The availability of this information for all births is the basis for drawing stratified samples and, ultimately, for generalizing results to the state's entire population of births. The availability of birth certificate data for all sampled women, whether they responded or not, is key to deriving nonresponse weights. For each respondent, the initial sampling weight is the reciprocal of the sampling fraction applied to the stratum. Sampling fractions in PRAMS range from 1 in 1 (for very low-birthweight strata in small states) to about 1 in 300 (for normal birthweight, non minority strata in populous states).
Corresponding sampling weights, then, would range from 1 to 300. Nonresponse adjustment factors attempt to compensate for the tendency of women having certain characteristics (such as being unmarried or having less education) to respond at lower rates than do women without these characteristics. Where multivariate analysis shows that these characteristics affected the propensity to respond in a particular stratum, the adjustment factor is the ratio of the sample size in that category to the number of respondents in the category. If analysis shows that no characteristic distinguishes respondents from nonrespondents, the adjustment factor is the ratio of the sample size in that stratum to the number of respondents in the stratum.
In the first case, each category so identified has an adjustment factor; in the second, there is a single factor for the whole stratum. The rationale for applying nonresponse weights is the assumption that nonrespondents would have provided similar answers, on average, to respondents' answers for that stratum and adjustment category. To ensure that cells with few respondents are not distorted by a few women's answers, small categories are collapsed until each cell contains at least 25 respondents. The magnitude of the adjustment for nonresponse depends on the response rate for a category. If 80% (4/5) of the women in a category respond, the nonresponse weight is 1.25 (5/4). Categories with lower response rates have higher nonresponse weights.
The frame noncoverage weights were derived by comparing frame files for a year of births with the calendar year birth tape that states provided to CDC. Omitted records are usually due to late processing and are evenly scattered across the state, but sometimes they are clustered by particular hospitals or counties or even by time of the year. The effect of the noncoverage weights is to bring totals estimated from sample data in line with known totals from the birth tape. In mail/telephone surveillance, the magnitude of noncoverage is small (typically from 1% to 5%), so the adjustment factor for noncoverage is not much greater than 1.
To derive these weights, PRAMS staff carried out such a frame omission study to look for problems that occurred during frame construction for all states. Multiplying together the sampling, nonresponse, and noncoverage components of the weight yields the analysis weight. The analysis weight can be interpreted as the number of women in the population who have characteristics similar to those of the respondent. Weighted results in SC PRAMS reports are produced with SUDAAN. SUDAAN is used for analyzing PRAMS data because it accounts for the complex sampling designs that states employ. It uses first-order Taylor series approximations to calculate appropriate standard errors for the estimates it produces.
This section presents an overview of:
1) The sampling approach used in PRAMS.
2) The derivation of the analysis weights applied to the weighted dataset.
3) the distribution of response rates.
4) A description of the method used to calculate and interpret any confidence intervals presented.
The SC PRAMS project utilizes a systematic stratified sampling strategy that takes birthweight into consideration. This is the most appropriate and efficient sampling strategy when the goal is to ensure large sample strata from groups that occur at low frequency in the total population (e.g. very low birthweight (VLBW) infants (<1,500 grams) account for close to 2% of the total live births). Therefore, SC PRAMS has sampled women having live MLBW (moderately low birthweight infants (1,500-2,499 grams) and VLBW infants at a higher rate than women having normal birthweight infants (2500 grams or more). Over-sampling of the low frequency strata ensures that reliable estimates of statistics can be presented separately for women having LBW infants.
|Very low birthweight (<1500 grams)|
|Moderately low birthweight (1500-2499 grams)|
|Normal birthweight (>=2500 grams)|
Computation of Analysis Weights
The SC PRAMS survey is designed to provide statewide estimates of the characteristics of women delivering live infants -- for example, the percentages of mothers who initiated prenatal care in the first, second, and third trimester, respectively; or the percentage of mothers who drank alcohol three months before they got pregnant or during the last trimester. To make such estimates each respondent must be assigned an “analysis weight.” This is a multiplier that is the number of women in the population she represents after adjustments for survey design, non-response and frame coverage. The analysis weight is the product of three sub-components weights. Each sub-component weight accounts for a different factor. The first sub-component adjusts for the sample design, the second adjusts for non-response, and the third for omissions in the sampling frame (i.e. non-coverage of the sampling frame). The PRAMS staff received technical assistance from the CDC to develop and compute the analysis weights applied in the weighted dataset. The three steps involved in deriving the analysis weights are described next:
A. Adjustment for sample design: the first component is called the sampling weight and it corresponds to the reciprocal of the sampling fraction (shown on the previous page). For example, in the moderately low birthweight stratum, 2 out of every 13 mothers is sampled. The sampling weight applied to respondents in this particular stratum is 6.5 (13/2).
B. Adjustment for non-response: the second component is called the unit non-response weight. The failure of the mother in the sample to complete a questionnaire is called unit non-response. Response adjustment cells were identified from extensive analysis of maternal characteristics affecting response rate within each birthweight stratum. The important maternal characteristics affecting response rates were maternal age, education, marital status, and race. The unit non-response weight is the product of the sampling weight times the inverse of the response rate specific to that response adjustment cell (based on maternal age, race, education, and marital status). For example, if the weight for a respondent from the moderately low birthweight stratum was 6.5 and that respondent was in a response adjustment cell with a 65 percent response rate, then the non-response adjusted weight for that respondent would be 10 (6.5/0.65). The lower the response rate for a particular r response adjustment cell, the larger the adjustment for non-response. Computation of the unit response weights rests on the assumption that within a stratum and non-response adjustment. cell, the average of the answers of the respondents is the same as the average of the answers of the non-respondents. As a rule, if there are fewer than 25 respondents in any response adjustment cell, the response adjustment category is combined with one or more other response categories until all response categories have at least 25 respondents. This ensures enough respondents in each category so that the average of their responses is not unduly influenced by a few women who participated.
C. Adjustment for omissions in the sampling frame (incomplete frame): The third component is called the sampling frame non-coverage weight. This weight adjusts for women whose live births were not included in the sampling frame (birth registry). The South Carolina birth registry system in the Office of Vital Records is efficient and expedient; therefore, the corrected sampling frame is very similar to the original frame. As a result, the ratio is close to 1 and the adjustment for non-coverage is very small.
Distribution of Response Rates
The weighted response rate for each sampling stratum that the CDC considers to be the threshold for epidemiologic validity is 70%. In 2007, the weighted response rates by sampling strata achieved by South Carolina PRAMS are as follows:
|Very low birthweight (<1500 grams)|
|Moderately low birthweight (1500-2499 grams)|
|Normal birthweight (>=2500 grams)|
The overall weighted response rate for 2007 South Carolina PRAMS data is 68%. Because this response rate is below the 70% threshold, any results obtained using 2007 SC PRAMS data should be interpreted with caution.
Calculation and Interpretation of Confidence Intervals
Background on Confidence Interval Calculation
In general, under the assumption of normality, a 95% confidence interval is constructed around a point estimate (e.g. a mean, percentage, proportion, regression parameter estimate, etc.) by using the following formula: CI = estimate ± (1.96*SE), where SE is the standard error of the estimate. Here SE = s/sqrt(n), where s is the standard deviation and n is the number of observations in the sample. Note that s2 is the variance of the point estimate. Generally the point estimate and the variance, s2, are calculated and then the confidence interval is constructed.
Variance Estimation for Complex Sampling Designs
When analyzing data from a survey that employs a complex sampling design, such as PRAMS, this variance needed to construct the confidence interval can not be correctly calculated using the standard formula and, therefore, must be estimated. The CDC recommends the use of the Taylor series linearization variance estimation technique. This technique does not directly estimate the variance, but rather uses a Taylor series expansion to estimate the point estimate and then calculates the variance of this estimated point estimate. This variance estimation method is appropriate for several sampling designs including stratified random sampling without replacement, which is the design used by SC PRAMS.
Confidence Interval Transformation
In addition to using an estimated variance, confidence intervals for proportions, percentages, etc. using PRAMS data are calculated using a logit transformation. This logit transformation ensures that confidence intervals are between 0 and 1 for proportions and between 0% and 100% for percentages. This logit transformation also causes the confidence interval to be asymmetric.
SC PRAMS uses the confidence interval interpretation that is recommended by the CDC. That is, “if the same sampling procedure were used to obtain many samples, and if a 95% confidence interval for a given population parameter were calculated from each sample, then 95% of the confidence intervals would actually include the value of the population parameter; 5% would not. … A ‘practical’ interpretation of the 95% confidence interval is it includes the value of the population parameter with a probability of 0.95.”1
Statistical Software Packages
Correct confidence intervals that account for the complex sampling scheme used to gather PRAMS data and that use appropriate transformations can be constructed using SUDAAN software as well as the SURVEY procedures in SAS and the SVY procedures in STATA. SC PRAMS usually conducts analyses using SAS-callable SUDAAN.
1Guidelines for State Analyses of PRAMS Data. (Inside PRAMS website requires username and password.) Accessed 03/03/2008.