National Incidence Studies of Missing, Abducted, Runaway, and Thrownaway Children (NISMART), 1999 (ICPSR 4566)

Time Period(s) View help for Time Period(s)

Date of Collection View help for Date of Collection

Sample View help for Sample

Parts 1-2: A list-assisted RDD (Random Digit Dial) methodology was used to select a nationally representative sample of telephone households from the GENESYS Sampling System frame. Typically, a list-assisted RDD design is a one-stage random selection process resulting in equal probabilities of selection (EPSEM) design and no clustering, with each household treated as a Primary Sampling Unit (PSU). A detailed description of the sampling design is in HOUSEHOLD SURVEY METHODS TECHNICAL REPORT which is included in this data collection. Part 3: A stratified, two-stage sample design was applied. In the first stage, 30 Primary Sampling Units (PSUs) were sampled from a listing of the national universe of counties and groups of small, adjacent counties. The PSUs were selected with probability proportionate to the number of children (aged 0 to 17) according to the 1990 Census. Within the 30 PSUs, a total of 922 eligible facilities were identified. This frame was stratified by type of facility, and 75 facilities were sampled with probability proportionate to size (PPS), using weighted capacity as the measure of size. Parts 4-6: Nationally representative sample of 400 counties. Counties were selected with probabilities proportional to size.

Time Method View help for Time Method

Universe View help for Universe

Parts 1-2: Households with at least one child aged 18 or younger who had lived in the house for two consecutive weeks in the 12 months prior to screening. Part 3: All group homes, juvenile detention centers, residential treatment centers, and runaway and homeless youth shelters serving the selected counties. Parts 4-6: All law enforcement agencies serving the selected counties.

Unit(s) of Observation View help for Unit(s) of Observation

Data Type(s) View help for Data Type(s)

Mode of Data Collection View help for Mode of Data Collection

Response Rates View help for Response Rates

Part 1: 35 percent. Part 2: 61 percent. Part 3: The JFS response rate for the institution-level interview was 100 percent. A total of 74 facility-level interviews were completed from the 75 sampled facilities. Three of the sampled facilities were found to be non-operational (hence ineligible), while a fourth sampled facility actually consisted of three separate juvenile residential facilities, all of which were included in the study. Thus, all in-scope facilities participated in the facility-level interviews. The JFS response rate for the episode-level interview was 93 percent. A total of 116 episode-level interviews were targeted. Episode-level interviews were successfully completed for 108 of these targeted episodes, yielding a 93 percent completion rate. Parts 4-6: Phase 1 mail survey -- 91 percent, and Phase 2 followup telephone interview -- 99.3 percent. Overall -- 90.6 percent.

Original Release Date View help for Original Release Date

2007-07-19

Version History View help for Version History

Weight View help for Weight

Parts 1-2: Survey weights were developed to compensate for the higher probability of selection of multiple-line households for children who lived in more than one household during the 12 months prior to interview, and to adjust for nonresponse and undercoverage of nontelephone households. The calculation of the RDD sample weights for the Household Survey was done sequentially and consisted of four main steps that included computing a base weight and various adjustments to it. Details about the weighting procedures for the adult and youth data are available in the NISMART-2 HOUSEHOLD SURVEY METHODOLOGY TECHNICAL REPORT which is included in the data collection. Part 3: In order to estimate the national prevalence of facility runaways with known precision, it was necessary to compute facility weights, episode weights, and replicate weights. The facility base weight was the inverse of the unconditional probability of selecting the facility. This was computed as the product of the PSU base weight and the facility base weight. Because there was no nonresponse at the facility level, the facility base weights were also the final facility-level weights. Because data on individual runaways were obtained through the episode-level interviews, the episode weights were equivalent to youth-level weights. These were developed in two steps. First the episode base weight was computed as the inverse of the conditional probability of selecting an episode, taking into account both the probability of selecting the facility that reported the episode and the probability of selecting the episode in question from all episodes reported by the facility. Second, because facilities could not provide details on eight of the 116 episodes they had identified, nonresponse adjustments were computed within facility categories to correct for the absence of these interviews in the episode database. The fact that the study used a multi-stage sampling design (sampling PSUs and then facilities within PSUs) meant that special procedures were required to produce appropriate standard errors for the study estimates. Most standard statistical packages do not produce correct variances because they treat the data as if all the elements in a sample had an equal and independent chance of selection. To allow computation of appropriate standard errors, the replication method was used to develop replicate weights for both the facility and the episode-level data. Thus, the final step in the weighting process was to develop replicate weights. Parts 4-6: Three types of weights were computed for sampled LES cases: agency weights, case weights, and replicate weights. The agency weight is a function of the probability of having selected the agency. Since all law enforcement agencies from the sampled counties were included in the sample, this was the same as the probability of having selected the county. The agency weights were also adjusted for nonresponse and refusals, thus letting the participating agencies represent the additional agencies that would otherwise be unrepresented because of the agencies that did not participate. Case weights took account of the probability of having sampled the case, which was generally the same as the probability of having sampled the agency from which the case originated. The case weights were used to correct for the loss of the two case-level refusals. The final step in the weighting process was the development of replicate weights. The complex sampling design required special procedures in order to compute the precision of the study estimates and conduct statistical tests. The replication approach was used with the LES data. The replication method required the development of replicate weights. While the replication method is not the only way to compute the precision of estimates from complex survey samples, replication methods have been shown to have several advantages even when other methods can be applied. The replicate samples are subsets of the full sample created to mirror the design of the full sample. The technique computes variance by measuring the variability among replicates of the full sample. When a sample includes more than 20 percent of the population, the method of computing replicates must be adjusted. As noted above, it was anticipated that the LES methodology would yield a sample of stereotypical abductions that would reflect a large percentage of the estimated total in the nation. This indeed turned out to be the case. Because of this, the replicates were adjusted with a "finite population factor" in order to avoid overestimating variances Case sampling was used in a few cases to reduce response burden in some large agencies in order to make it feasible for the agency to participate in the study. In these instances, case weights ensured that the sample cases represented all cases in the agency.