Appendix 8: Instrument Rosters

A roster may be thought of as a list: for example, a list of household members, a list of employers, or a list of children. A respondent with two children will have data on the first two lines of the child list, or child roster. A respondent with four employers will have information on the first four lines of the employer roster. In addition to the name of the person or item (which is not released to the public), the roster contains other basic information, such as the age, race, and labor force status of household members or the start date and stop date for each employer.

In the paper-and-pencil interviews (PAPI) of older NLS cohorts, the questionnaires included a chart or grid listing this type of information, like the one shown in Figure 1 below. For example, in the household roster grid, each household member's name was entered in a separate row. The interviewer asked the respondent for each member's date of birth, enrollment status, employment status, etc., filling in the answers in the appropriate column. This completed household roster contained all the pertinent information about household residents, and researchers could easily use the variables based on this roster to examine characteristics of household members.

When the NLS surveys changed to computer-assisted personal interviewing (CAPI), rosters became a very important way of organizing information during the interview. Instead of using an actual grid, however, CAPI questionnaires include a series of questions that gather the same types of information that would have been included in the grid in a paper-and-pencil interview. The computer then moves the answers to these questions into a grid, creating a roster from the information.

After the roster is created, it can be used to guide subsequent portions of the interview. For example, during the interview the NLSY97 questionnaire gathers the names, dates of attendance, and level of school (secondary school or college) for each of the respondent's schools and organizes them into a roster. The rest of the school section then asks questions about the first school on the roster, followed by questions about the second school, then the third, and so on. The information about the level of the school determines whether the respondent is asked questions that apply to high school or college.

The information from the roster is also presented in the data set as an organized list of data, so that these variables are easy for researchers to access. To the user, the school roster appears as a consolidated block of variables that contains key information such as dates of enrollment, an identification number for the school, and variables indicating the type (private or public) and level (junior high, high school, college) of the school. For example, the variables in the round 2 school roster are listed in Figure 2, along with their reference numbers. Thus, rosters are a way of organizing information both for researchers and for the actual interview so that questions are asked in a logical manner.

This section includes the Creation and use of the employer roster example.

This section outlines the process used during the interview to create a roster. Rosters may include data from both previous interviews and the current interview. After the roster is created and sorted, it can be used to guide the rest of the interview. Figure 3 provides a pictorial overview of the creation of a roster.

Data from previous interviews

As shown in the figure, creation of a roster for the current round often begins with information found in the roster from the previous round. The appropriate respondent-specific data are saved on the interviewer's laptop before he or she administers the survey. When the interview gets to a point where roster information is collected, the data from the previous round's roster are often used as the base for the current roster. The respondent verifies and updates the information. If no changes have occurred since the last interview-for example, if exactly the same people live in the respondent's household-then the current round's roster will be the same as the one from the previous round.

For example, the interviewer reads a list of all of the people on the household roster from the last interview. The respondent first states whether any of those people have moved out of the household and then reports new household members. If any members remain from the previous year, their information-date of birth, sex, race/ethnicity, etc.-is carried over from the previous interview, and any missing data are collected. This method is more efficient than asking the respondent to report all household members every year.

Raw data collection

After the respondent and interviewer review and update the roster from the previous round, the survey collects current information. For example, new people might have moved into the household, so the interviewer asks the respondent about their characteristics. At this point, the respondent is done answering questions that will fill up the data grid on a particular topic.

Roster creation and roster sort

Using the updated roster from the previous round and the new raw data just collected, the computer creates a new roster for the current round. For example, the employer roster contains the following information for each job: a unique identification number for the employer, employment dates, whether the job was current at the interview date, whether the job was in the military, and whether the job was an internship. If the respondent had held the job at the time of the previous interview, the start date and employer identification number are carried over from the old roster, and the other information is taken from the questions at the beginning of the employment section for the current year. Similarly, the household roster contains information from the previous interview about household members reported at that time and data from the current interview about new household members.

In some cases, the computer also sorts the roster and puts the items in order based on a specified variable. For example, in the round 1 household roster, all youths in the age range of the NLSY97 cohort were listed first, and then all other household members were listed from oldest to youngest. The employer roster is sorted by job end date so that the most recent jobs are listed first.

Roster use in the interview

Finally, the roster is used to determine the order in which the other questions about each topic are asked. In most cases, the survey collects far more information than is stored in the actual roster, and the answers to these questions remain outside the roster as raw data. So that the interview makes sense to the respondent, these additional questions are asked about the people or things on the roster in the order that the people or things are listed.

For example, the respondent first answers questions about industry, occupation, rate of pay, etc., for the first employer listed on the roster. The same questions are then asked about the second job, then the third job, and so on. Similarly, the first set of questions about household members refers to the first person listed on the roster. When all of those questions have been answered, the same questions are asked about the second person, the third person, etc.

This section includes the Use of the employer roster in analysis example.

The data set is organized so that rosters can easily be found and used in research. Because rosters present key pieces of information in a structured format, they are the best place to obtain that information. All variables found on rosters have "Roster Item" as their main area of interest. Each roster has a unique name that serves as the beginning of the question name for all variables on the roster; the same name appears at the beginning of the variable title for each item on the roster. Different rosters have been used in different rounds, depending on the topics included in the interview and the type of information collected. The roster names and question names are shown in Figure 4.

Using rosters in single-round analyses

When looking at the data set, users will notice that many questions are repeated for each person or thing on the roster, and the titles for these repeated questions include a number. This number indicates the line on the roster that corresponds to the person or item being described in that variable. For example, the question "Self-Employed Business/Industry Job 02" indicates the industry of the second job listed on the respondent's self-employment roster. The researcher may then want to examine information such as the respondent's start and stop dates or rate of pay for that job. To find this information, he or she can then look at the data for those items contained in the roster for job #02, or the self-employment job that is on the second line of the roster. For all other questions asked after the roster was created in that same survey year, job #02 will refer to the same self-employment job.

Users should be aware that, in some cases, the information contained in the rosters actually appears in the data set more than once. As Figure 1 suggests, data may first be included at the point in the interview when the information was actually collected. For example, the round 1 screener question SE-28 asked the household informant for the date of birth of each household member. After all the raw data had been gathered, the computer sorted all the answers and created the household roster. At this point the date of birth information is also located in the round 1 roster variables named HHI2_DOB. In the case of the round 1 household roster, both the raw data and roster items are included in the data set.

In other cases, the raw answers may be blanked out of the public use data set. If a reference number is not listed for a given question in the questionnaire, then that raw data item may only be represented in roster form. For example, answers to the raw data questions used to create the employer roster are blanked out and do not appear in the data set. In the printed questionnaire, these questions have no reference numbers. However, all of the data collected in these questions (except for confidential information like the name of the employer) appears in the employer roster.

Using rosters from more than one round

Because the NLSY97 is a longitudinal survey, researchers often want to link data across survey rounds. However, household residents, jobs, and so on may move around on the roster in different interviews. That is, a father who was listed third on the roster in round 1 might move to position 2 or 4 in round 2. The unique identification numbers (UIDs) are the key to finding the same person or thing in different rounds. Most of the rosters contain variables assigning a unique number to each person or thing listed. This number never changes and can be used to link roster items across rounds. In some cases, it also makes it possible to link people between two different rosters in the same survey. For example, beginning in round 2 the unique ID listed for a child on the biological children roster is the same one assigned to that child on the household roster. Researchers can therefore examine data on both rosters about the same child.

An additional feature of most unique ID numbers is that they incorporate an indicator of the round in which the person or item was first reported. For example, IDs of roster items reported in round 1 may begin with "1" or "97," while those first reported in round 2 begin with "2" or "98." (Beginning with round 3, 4-digit years are used so that IDs begin with "1999" rather than just "99.") UIDs for people on the household roster are constructed in a slightly different manner; researchers should refer to Household Composition for more information.

Example: Use of the employer roster in analysis

Continuing the above example, this section explains how to use rosters in data analyses. Although the employer roster is used in the example, most aspects apply to other NLSY97 rosters as well. Emma's information, as organized in the employer rosters, can be used to examine the characteristics of her jobs at the date of each interview or over time. This example focuses primarily on the round 2 employer roster, but use of the roster is similar in each subsequent round.

As described above, Emma worked for Peel's Store and Steed's Diner during the period between the round 1 and round 2 interviews. Information about these employers was sorted and a roster constructed with the most recent employer appearing first. A researcher using these data would need to be aware of the impact of roster construction.

Because the roster is sorted and employers reported in different rounds may be mixed, variables with "Employer #01" in the title do not necessarily refer to employer number 9701, 9801, etc. The #01 refers solely to the order of the job as listed on the current year's roster. The unique identification numbers provide a crosswalk between the two systems of identification. The UIDs also allow users to link employers across survey rounds and to identify the round in which an employer was first reported.

For example, Emma's value for the round 2 variable R24761., "YEMP, Employer 02 Unique ID (Ros Item)," would be 9703-Peel's Store. The user can identify this as an ID assigned in round 1 because it starts with "97," and look at the round 1 UID variables (R05311.-R05317.) to match the employer. In Emma's case, the variable for employer #01 in round 1 would have UID 9703. Therefore, the researcher knows that information about employer #01 in round 1 refers to the same job as variables about employer #02 in round 2. The variables from the two rounds can then be compared to determine if there were any changes in characteristics such as hours worked, rate of pay, occupation, etc.

The roster line numbers and UID variables in the event history data work in the same way. For example, a researcher might want to know Emma's employment status in the first and last week of 1998. In the first week of 1998 (variable EMP_STATUS.01.98), Emma was working at her parents' store, so the status variable would have a value of 9703. Using this UID, researchers can link that job to all of the other information collected during the interview. For the last week of 1998 (variable EMP_STATUS.01.98), when Emma was working at Steed's Diner, the status variable would have a value of 9801. The second set of event history variables, the start and stop dates of each job, uses the roster line numbers. For these variables, the number in the variable title refers to the same job as in the main data set. For example, in the main round 2 data Peel's Store is job #02. The start and stop dates for Peel's Store in the event history data (variables EMP_START_WEEK.02 and EMP_END_WEEK.02) will also have #02 in the variable title.

Household Information

PREFIX = "HHI" (e.g., HHI_AGE.01); in round 1 only prefix was "HHI2" (e.g., HHI2_AGE.01)

NONHHI Roster (Non-resident relative enumeration)

PREFIX = "NONHHI" (e.g., NONHHI_AGE.01)

PARTNERS Roster and CUMPARTNERS Roster (Youth partners)

PREFIX = "PARTNERS" (e.g., PARTNERS_xxxxx.01) and ("CUMPARTNERS" (e.g., CUMPARTNERS_xxxxx.01)

Note: A PARTNERS Roster is created for each survey round. Then, the CUMPARTNERS Roster is created as a collapsed roster which contains data from the most recent round in which each respondent was interviewed. Both rosters are represented here as the items on the rosters are very similar.

BIOChild Roster (Youth's children)

PREFIX = "BIOCHILD" (e.g., BIOCHILD_BDATE.01)

BIOADOPTChild Roster (Respondent's children)

PREFIX = "BIOADOPTCHILD" (e.g., BIOADOPTCHILD_SEX.O1)

OTHER PARENT Roster (Other parent of respondent's children)

PREFIX = "OTHERPARENTS" (e.g., OTHERPARENTS_AGE.01)

Note: This is a collapsed roster and contains data from more than one survey round. It includes information for all respondents, regardless of whether they were actually interviewed in the most recent round. In Investigator, the survey year is listed as "XRND".

NEWSCHOOL Roster (Respondent's schools)

PREFIX = "NEWSCHOOL" (e.g., NEWSCHOOL_PERIODS.01)

YEMP Roster (Youth's employers)

PREFIX = "YEMP" (e.g., YEMP_CURFLAG.01)

FREELANCE Roster (Respondent's freelance jobs)

PREFIX = "FREELANCE" (e.g., FREELANCE_JOBS-COD.01)

TRAINING Roster (Respondent's training programs)

PREFIX = "TRAINING" (e.g., TRAINING_UID.01)

Additional roster variables

A handful of other variables have been listed as roster variables in various survey rounds. These special variables are summarized below.

In round 1, a number of rosters were used to organize information from the Screener, Household Roster, and Non-Resident Roster Questionnaire. Some of this information was then transferred into the parent and youth interviews for verification and for use in determining question paths. Figure 5 identifies the key rosters in the round 1 survey and shows how they were used in different parts of the survey.

The YOUTH, PARHHI, and PARYOUTH rosters appeared only in the round 1 survey. The contents of these special rosters are listed following Figure 5. Other rosters in the figure also appeared in subsequent rounds and are listed in the previous section.

YOUTH Roster (Youth respondent information)

PREFIX = "YOUTH" (e.g., YOUTH_ADOPDADID.01)

PARHHI Roster (Household information in parent interview)

PREFIX = "PARHHI" (e.g., PARHHI_AGE.01)

PARYOUTH Roster (Youth information for parent interview

PREFIX = "PARYOUTH" (e.g., PARYOUTH_AGE.01)