Nonresponse

All surveys experience some degree of nonresponse. As discussed in Chapter 6 of the SIPP Users' Guide, in a longitudinal survey such as SIPP, as the number of waves increases, nonresponse may result in a corresponding increase in bias. Since nonrespondents may differ from respondents in terms of the variables collected in the survey, the occurrence of nonresponse gives rise to concerns about bias in the survey results. Weighting adjustments are made in an attempt to reduce or eliminate bias (Chapter 8 of the SIPP Users' Guide), but concerns about nonresponse bias remain.

The rate of sample loss ¹² in SIPP generally declines from one wave to the next. The total number of sample members lost, also known as total sample attrition, always increases over time.

Wave 1 nonresponse rates for SIPP have been about 7.7 percent.¹³ There is usually a sizable sample loss at Wave 2, with a lower rate of additional attrition occurring at each subsequent wave. Prior to the 1992 Panel, SIPP lost roughly 20 percent of the original sample by the panel's completion. The sample loss rate for the 1996 Panel was 35.5 percent by the end of the 12th, or final, wave. Chapter 6 in this volume and the SIPP Quality Profile provide more detailed discussions of the implications of nonresponse for data quality. SIPP deals with the various types of nonresponse by weighting adjustments or imputation (Chapters 8 and 4 of the SIPP Users' Guide). Table 2-5 shows cumulative loss rates for two types of nonresponse, discussed below.

The Census Bureau distinguishes between household and person nonresponse. Household nonresponse occurs either when the interviewer cannot locate the household or the when interviewer locates the household but cannot interview any adult household members. Person- level nonresponse occurs when at least one person in the household is interviewed and at least one other person is not.usually because that person refuses to answer the questions, or is unavailable and no proxy is taken. The Census Bureau categorizes household nonresponse as Types A and D (detailed definitions and discussion of rates follow),¹⁴ and person-level nonresponse as Type Z.

Household Nonresponse

Type A household nonresponse occurs when the interviewer finds the household.s address, but obtains no interviews. Those households contain people eligible for SIPP interviews, but every eligible member of the household is a noninterview. Examples of Type A nonresponse include the following:

The interviewer finds no one at home despite repeated visits.
All eligible household members are away during the entire interview period (e.g., an extended vacation).
Household members refuse to participate in the survey.
The interviewer cannot reach the housing unit because of impassable roads, such as from a natural disaster.
Interviews cannot be taken because of serious illness or death in the household.

When this type of household nonresponse occurs in Wave 1, SIPP makes no attempt to interview the household members at subsequent waves. For Type A nonresponse that occurs in subsequent waves, however, interviewers try to obtain interviews on the following wave. New Type A noninterviews represent the first time a Type A household nonresponse occurred. Old Type A nonresponse represents unsuccessful attempts to convert a Type A noninterview from the previous wave. Two consecutive Type A noninterviews render the case ineligible for interviews at the following wave.¹⁵

Table 2-5. Household Noninterview and Sample Loss Rates: 1990–1996 Panels

Wave	1990 Panel			1991 Panel			1992 Panel			1993 Panel			1996 Panel
	Type A	Type D	Loss	Type A	Type D	Loss	Type A	Type D	Loss	Type A	Type D	Loss	Type A	Type D	Loss
1	7.3	–	7.3	8.4	–	8.4	9.3	–	9.3	8.9	–	8.9	8.4	–	8.4
2	10.9	1.5	12.6	12.3	1.5	13.9	12.8	1.7	14.6	12.4	1.7	14.2	13.1	1.3	14.5
3	11.5	2.6	14.4	13.1	2.7	16.1	13.1	2.8	16.4	12.9	2.9	16.2	15.6	1.9	17.8
4	12.5	3.4	16.5	13.6	3.6	17.7	13.8	3.6	18.0	13.9	3.8	18.2	17.6	3.1	20.9
5	13.6	4.6	18.8	14.5	4.2	19.3	14.9	4.7	20.3	14.9	4.7	20.2	20.4	3.8	24.6
6	14.1	5.3	20.2	14.4	5.1	20.3	15.3	5.4	21.6	15.9	5.5	22.2	22.2	4.4	27.4
7	14.3	5.9	21.1	14.7	5.6	21.0	16.0	5.9	23.0	17.2	6.2	24.3	23.8	4.8	29.9
8	14.4	5.9	21.3	14.5	5.9	21.4	16.9	6.7	24.7	17.5	6.9	25.5	24.2	5.4	31.3
9	–	–	–	–	–	–	17.7	7.3	26.2	18.2	7.5	26.9	25.0	5.6	32.8
10	–	–	–	–	–	–	17.5	7.6	26.6	–	–	–	26.1	6.0	34.0
11	–	–	–	–	–	–	–	–	–	–	–	–	25.5	6.2	35.1
12	–	–	–	–	–	–	–	–	–	–	–	–	–	6.2	35.5

Note: The sample loss rate is the cumulative noninterview rate adjusted for unobserved growth in the Type A noninterview units (created by splits). Source: SIPP Quality Profile, 3rd Ed. (U.S. Census Bureau, 1998a).

Type D household nonresponse concerns original sample members who move to an unknown or uninterviewable address; it applies only to Wave 2 and beyond. Those noninterviews occur when a household or some members of a household are living at an unknown new address or at an address located more than 100 miles from a SIPP sample area and cannot be contacted by telephone.¹⁶ For the 1996 Panel, Type D noninterviews are attempted three times before they are dropped.

Person Nonresponse

There are two forms of person-level, or Type Z, nonresponse. The first applies to those instances in which a sample person was in the household during part (or all) of the reference period and was part of the household on the date of the interview but refused to answer, or was not available for the interview and a proxy interview was not obtained. The second form of Type Z noninterview occurs when a person was part of the household during part of the 4-month reference period but then moved and was no longer a household member on the date of the interview.¹⁷ While household nonresponse is usually handled by weighting adjustments, Type Z cases are handled by imputation (i.e., they are matched to donors, and data from the donor case are substituted for the missing interview. (see discussion of imputation and weighting in Chapters 4 and 8 of the SIPP Users' Guide). Nearly half of SIPP Type Z nonrespondents are not interviewed at any of the waves.

Item Nonresponse

Item nonresponse is an additional source of missing data; it occurs when a respondent does not answer one or more questions, even though most of the questionnaire is completed. Respondents might refuse to answer a particular question or set of questions. Sometimes, item nonresponse occurs when respondents do not have the information requested.¹⁸ Although interviewers are trained to attempt to persuade respondents to answer all applicable questions, and will call back if a respondent can provide data at a later time, those efforts are not always successful. Item nonresponse can also result from the postinterview data editing process when respondents provide inconsistent information or when an interviewer incorrectly records a response. In many cases, the Census Bureau handles item nonresponse by imputation, that is, by assigning values for the missing items (Chapter 4 of the SIPP Users' Guide).

__________
¹² The accumulation of cases that are no longer being interviewed because of as yet unrecovered refusals or as yet unfound movers.
¹³ Nonresponse rates have not been stable, ranging from 6.70 percent for the 1984 through 1990 Panels to 8.48 percent for the 1991 through 1996 Panels.
¹⁴ The Census Bureau recognizes two other types of household noninterviews. Type B occurs in Wave 1 when the address unit is vacant or in some way unfit for residence; in subsequent waves, Type B occurs when people enter institutions. Type C occurs in Wave 1 when the housing unit has been demolished or converted to some other use; in subsequent waves, Type C occurs when all sample members in a household are outside the scope of the survey, e.g., deceased, living abroad, or living in armed forces barracks.
¹⁵ For each wave, the rate of Type A nonresponse is calculated by adding the number of Type A noninterviews for the wave to the number of Type A noninterviews dropped from the sample in prior waves and dividing that sum by the total of the number of interviewed households plus all Type A and Type D noninterviews.
¹⁶ For each wave, the rate of Type D nonresponse is calculated by adding the number of Type D noninterviews for the wave to the number of Type D noninterviews dropped from the sample in prior waves, and dividing that sum by the total of the number of interviewed households plus all Type A and Type D noninterviews.
¹⁷ If the person was an original sample member, information will be taken for the portion of the reference period in which he or she was still at the address, and an effort will be made to locate the person. If the person was not an original sample member, information will be taken for the portion of the reference period in which he or she was still at the address, after which the person will not be pursued.
¹⁸ The information provided may also be inconsistent with edit specifications, and the response is thus deleted during the processing stage. Or, interviewers may forget to ask for the information or record it incorrectly, resulting in an edit failure. See Chapter 4 of the SIPP Users' Guide on editing and imputation.

Sample Design

Organizing Principles

Selection of Sampling Units

Oversampling Principles

Identifying Sample Members

Following Rules

Interview Procedures
end of content rule

Page Last Modified: May 9, 2006

Census Bureau Links: Home · Search · Subjects A-Z · FAQs · Data Tools · Catalog · Census 2000 · Quality · Privacy Policy · Contact Us