Common use of Oversampling Clause in Contracts

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used for MEPS Panels 15 and 16 for subsampling among the NHIS respondents eligible for MEPS were somewhat different. For Panel 15 four domains (strata) were established in a hierarchical sequence. The first stratum contained households with Asians, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 15, all households in the Asian domain were sampled with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent for the “Other” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnaire.

Oversampling. Xxxxxxxxxxxx Oversampling was employed for selected some subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is will be discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “"oversampled” " are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “"not oversampled”". As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “"cost” " of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved achieve if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, NHIS Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches approach differed somewhat between Panels 6 and the sampling domains used for 7 as, beginning with Panel 7, MEPS Panels 15 has begun to oversample Asians and 16 for subsampling among those predicted to be poor. From the NHIS respondents households eligible for MEPS MEPS, reflecting the oversampling of Hispanics and blacks described above, two strata were somewhat differentconstructed for sampling purposes. For Panel 15 four domains (strata) were established in a hierarchical sequence. The first One stratum contained households with Asians, a second stratum contained households with Hispanics not assigned Asians and those "predicted to be poor" while the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and the fourth other stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 15, all All households in the Asian domain were sampled with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent for the “Other” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 percent. Within strata (domains) for both panels, responding NHIS households "Asian/Predicted Poor" stratum were selected with certainty while roughly 50% of the other stratum was selected for MEPS MEPS, using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (Because Hispanics and blacks had been oversampled for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selectionas described above, households with Hispanics and Blacks were also included at disproportionately high rates (oversampled) among the households selected at the 50% rate. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsamplingThus, for MEPS, households that were oversampled for MEPS in calendar year 2011 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispaniccontaining Hispanics, BlackBlacks, or Asian. AgainAsians, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had and those predicted to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairebe poor.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That iswhere Hispanics, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Blacks, and Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used for MEPS Panels 15 16 and 16 17 for subsampling among the NHIS respondents eligible for MEPS were somewhat differentdiffered only in that a “cancer” domain was used for Panel 16. For Panel 15 four 16 six domains (strata) were established in a hierarchical sequence. The first stratum contained households with Asians, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; , a second stratum contained households with Asians, a third stratum contained households with Hispanics not assigned to the second stratum, a fourth stratum contained households with Black members not assigned to strata two and three, the fifth stratum contained Other households characterized as “complete” for the NHIS; and the sixth stratum contained Other households characterized as “partial complete” for the NHIS. For Panel 1517, after eliminating the cancer domain, the same hierarchical ordering was used. In terms of sampling rates, for Panel 16 all households in the Asian domain cancer, Asian, Hispanic, and Black domains were sampled with certainty. The sampling rate was about 76 79 percent for the Hispanic domain, about 86 percent for the Black “Other complete” domain, and about 61 46 percent for the “OtherOther partial complete” domain. For Panel 16 17 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 51 percent while the “Other, partial complete” domain was sampled at a rate of about 46 40 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within the two strata involving “Others” (for both Panel 15, one stratum; for 16 and Panel 16, two strata17) the selection sample was selected with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2012 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnaire.

Oversampling. Xxxxxxxxxxxx Oversampling was employed for selected some subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is will be discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “"oversampled” " are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “"not oversampled”". As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “"cost” " of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved achieve if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches approach for both Panels 10 and 11 was the sampling domains used for MEPS Panels 15 and 16 for subsampling same. From among the NHIS respondents households eligible for MEPS MEPS, reflecting the oversampling of Hispanics and blacks described above, three strata were somewhat differentconstructed for sampling purposes for MEPS. For Panel 15 four domains (strata) were established in a hierarchical sequence. The first One stratum contained households with AsiansAsians and those "predicted to be poor", a second stratum contained households with Hispanics black members that were not assigned to among the households in the first stratum, a while the third stratum contained households with Black members not assigned to the first two strata, and the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 15, all All households in the Asian domain "Asian/Predicted Poor" stratum were sampled selected with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent black stratum was three-fourths. The sampling rate for the “Otherother” domain. For stratum was about 53 percent for Panel 16 the corresponding sampling rates 10 and 56 percent for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 percentPanel 11. Within strata (domains) for both panelsstrata, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”Because Hispanics had been oversampled for the NHIS, as described above, households with Hispanics were also included at disproportionately high rates (oversampled) among the selection was with equal probabilityhouseholds selected at the roughly 50% rate. Within strata involving “Others” (Thus, for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsamplingMEPS, households that were oversampled for MEPS in calendar year 2011 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispaniccontaining Hispanics, BlackBlacks, or Asian. AgainAsians, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had and those predicted to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairebe poor.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That iswhere Hispanics, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not Blacks, and Asians are oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used approach for MEPS Panels 15 19 and 16 20 used the same sampling domains (strata) for subsampling among the NHIS respondents eligible for MEPS were somewhat differentMEPS. For Panel 15 four Five domains (strata) were established in a hierarchical sequence. The : the first stratum domain contained households with Asians, a ; the second stratum domain contained households with Hispanics not assigned to the first stratum, a domain; the third stratum domain contained households with Black members not assigned to the first two strata, domains one and two; the fourth stratum domain contained all those remaining households (“Other” households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households ) characterized as “complete” for the NHIS; and Other households the fifth domain contained all remaining “Other” households, those characterized as “partial complete” for the NHIS. For In terms of sampling rates, for Panel 15, 19 all households in the Asian domain Asian, Hispanic, and Black domains were sampled with certainty. The sampling rate was about 76 66 percent for the Hispanic domain, about 86 percent for the Black “Other complete” domain, and about 61 42 percent for the “OtherOther partial complete” domain. For Panel 16 20 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 84 percent while the “Other, partial complete” domain was sampled at a rate of about 46 53 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within the two strata involving “Others” (for both Panel 15, one stratum; for 19 and Panel 16, two strata20) the selection sample was selected with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2015 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnaire.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used for MEPS Panels 14 and 15 and 16 for subsampling among the NHIS respondents eligible for MEPS were somewhat differentsimilar. For Panel 15 four The same sample domains were employed but with slightly different sampling rates. Four domains (strata) were established in a hierarchical sequence. The first stratum contained households with Asians, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and while the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 1514, all households in the Asian domain stratum were sampled with certainty. The sampling rate was about 87 percent for the Hispanic stratum, about 90 percent for the Black stratum, and about 72 percent for the “other” stratum. For Panel 15 again all households in the Asian stratum were sampled with certainty. The sampling rate was about 76 percent for the Hispanic domainstratum, about 86 percent for the Black domainstratum, and about 61 percent for the “Otherother” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 percentstratum. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2010 were those responding the households in the containing Hispanics, Blacks, and Asians, based on their NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairemembership.

Oversampling. Xxxxxxxxxxxx Oversampling was employed for selected some subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is will be discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “"oversampled” " are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “"not oversampled”". As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “"cost” " of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved achieve if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, NHIS Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches approach differed slightly between Panels 8 and the sampling domains used 9. For both panels, NHIS responding households containing Asians and those predicted to be poor were oversampled for MEPS. This practice began with MEPS Panels 15 and 16 Panel 8. However, in addition, for subsampling among Panel 9, NHIS responding households with black members were sampled at higher rates than all other except those with Asian members or those predicted to be poor. From the NHIS respondents households eligible for MEPS MEPS, reflecting the oversampling of Hispanics and blacks described above, two strata were somewhat differentconstructed for sampling purposes. For Panel 15 four domains (strata) were established in a hierarchical sequence. The first One stratum contained households with Asians, a second stratum contained households with Hispanics not assigned Asians and those "predicted to be poor" while the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and the fourth other stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 15, all All households in the Asian domain "Asian/Predicted Poor" stratum were sampled selected with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent for certainty while roughly two thirds of the “Otherblack” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate stratum and 50% of about 79 percent while the “Other, partial completeother” domain stratum was sampled at a rate of about 46 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS MEPS, using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (Because Hispanics had been oversampled for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selectionas described above, households with Hispanics were also included at disproportionately high rates (oversampled) among the households selected at the roughly 50% rate. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsamplingThus, for MEPS, households that were oversampled for MEPS in calendar year 2011 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispaniccontaining Hispanics, BlackBlacks, or Asian. AgainAsians, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had and those predicted to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairebe poor.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used for MEPS Panels 15 12 and 16 13 for subsampling among the NHIS respondents eligible for MEPS were was somewhat differentsimilar. For The sampling domains are the same for both panels and the subsampling rates are all high except for the “Other” domain for Panel 15 13. From among the NHIS households eligible for the Panel 12, four domains (strata) were established in a hierarchical sequencesequence but in essence there were only two sampling strata employed. The first stratum contained households with AsiansAsians and those “predicted to be poor”, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black black members in households not assigned to the first two strata, and while the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding All households was: households where an individual was identified in the NHIS as having cancer via “Asian/Predicted Poor” stratum were selected with certainty while the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” sampling rates for the NHIS; and Other households characterized as “partial complete” remaining three strata were essentially the same, roughly 90 percent. Panel 12 was the first year in which Hispanics formed a separate sample domain for the NHISMEPS. For Panel 1513 there were also in essence only two strata as the domains associated with Asians, the “predicted to be poor”, Hispanics, and Blacks all households in the Asian domain were sampled with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent for the “Otherother” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain stratum was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 57 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2008 were the households containing Hispanics, Blacks, Asians, and those responding households in the predicted to be poor based on their NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairemembership.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used for MEPS Panels 15 13 and 16 14 for subsampling among the NHIS respondents eligible for MEPS were somewhat slightly different. For From among the NHIS households eligible for the Panel 15 13, four domains (strata) were established in a hierarchical sequencesequence but in essence there were only two sampling strata employed. The first stratum contained households with AsiansAsians and those “predicted to be poor”, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black black members in households not assigned to the first two strata, and while the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding All households was: households where an individual was identified in the NHIS as having cancer via the NHIS strata of “Asian/Predicted Poor”, Hispanics, and Blacks were sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” with certainty. The sampling rate for the NHIS; and Other “other” stratum was about 57 percent. Similar to Panel 13, there were also four strata established for Panel 14, with the only difference that the first stratum only contained households characterized as with Asians. Those households “partial completepredicted to be poor” did not constitute a separate sampling domain for the NHISPanel 14. For Panel 15, all All households in the Asian domain stratum were sampled with certainty. The sampling rate was about 76 87 percent for the Hispanic domainHispanics stratum, about 86 90 percent for the Black domainBlacks stratum, and about 61 72 percent for the “Otherother” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 percent while the “Other, partial complete” domain was sampled at a rate of about 46 percentstratum. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within strata involving “Others” (for Panel 15, one stratum; for Panel 16, two strata) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2009 were those responding the households in the containing Hispanics, Blacks, and Asians, based on their NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairemembership.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That iswhere Hispanics, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not Blacks, and Asians are oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used approach for MEPS Panels 15 18 and 16 19 used the same sampling domains (strata) for subsampling among the NHIS respondents eligible for MEPS were somewhat differentMEPS. For Panel 15 four Five domains (strata) were established in a hierarchical sequence. The : the first stratum domain contained households with Asians, a ; the second stratum domain contained households with Hispanics not assigned to the first stratum, a domain; the third stratum domain contained households with Black members not assigned to the first two strata, domains one and two; the fourth stratum domain contained all those remaining households (“Other” households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households ) characterized as “complete” for the NHIS; and Other households the fifth domain contained all remaining “Other” households, those characterized as “partial complete” for the NHIS. For In terms of sampling rates, for Panel 15, 18 all households in the Asian domain Asian, Hispanic, and Black domains were sampled with certainty. The sampling rate was about 76 63 percent for the Hispanic domain, about 86 percent for the Black “Other complete” domain, and about 61 43 percent for the “OtherOther partial complete” domain. For Panel 16 19 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 66 percent while the “Other, partial complete” domain was sampled at a rate of about 46 42 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within the two strata involving “Others” (for both Panel 15, one stratum; for 18 and Panel 16, two strata19) the selection sample was selected with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2014 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnaire.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups Oversampling is a feature of policy-level interest the MEPS sample design, helping to help increase the precision of estimates associated with members for some subgroups of those subgroupsinterest. Before going into detailsdetails related to MEPS, the concept of oversampling is will be discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for specific subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates an oversampled subgroup comprises a higher proportion of the sample than it represents in the samplegeneral population. Sample weights help ensure that population estimates account for this are not distorted by a disproportionate contribution from oversampled subgroups, as the base . Base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done implemented to increase the sample sizes and thus improve the precision of survey estimates for that particular subgroupsubgroups of the population. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For The oversampling of Hispanic, Black, and Asian households for the NHIS carries over to MEPS through the set of NHIS responding households eligible for sample selection for MEPS, some of . In the oversampling was achieved through its linkage to NHIS under the NHIS. For the earlier old sample design of the NHISutilized through 2005, Hispanic households were oversampled at a rate of roughly 2 to 1. That is, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not oversampled. The NHIS oversampling rate for black Black households under the old design was roughly 1.5 to 1. For Under the new NHIS sample design Asians employed through 2015 (which is the sample design applicable for MEPS Panels 20 and 21), Asians, as well as Hispanics and Blacks, are also oversampled. The estimated overall average oversampling rates associated with each of for the three minorities minority groups have not yet been reported. The oversampling approaches For both Panel 20 and the sampling domains used for MEPS Panels 15 and 16 for subsampling among the NHIS respondents eligible for MEPS were somewhat different. For Panel 15 four domains (strata) were established in a hierarchical sequence. The first stratum contained households with Asians, a second stratum contained households with Hispanics not assigned to the first stratum, a third stratum contained households with Black members not assigned to the first two strata, and the fourth stratum contained all remaining households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households characterized as “complete” for the NHIS; and Other households characterized as “partial complete” for the NHIS. For Panel 1521, all households in the Asian domain Asian, Hispanic, and Black domains were sampled with certainty. The sampling rate was about 76 percent for the Hispanic domain, about 86 percent for the Black domain, and about 61 percent for the “Other” domain. For Panel 16 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 certainty (i.e., all households assigned to those domains were included in the MEPS). The For Panel 20, the “Other, complete” domain was sampled at a rate of about 79 84 percent while the “Other, partial complete” domain was sampled at a rate of about 46 53 percent. For Panel 21, the corresponding sampling rates for the “Other, complete” domain and the “Other, partial complete” domain were about 81 percent and 49 percent, respectively. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For the “non-Other” strata households were all but those strata involving “Others”, the selection was selected with equal probabilitycertainty. Within strata involving “Others” (two strata for Panel 15, one stratum; for Panel 16, two strataboth panels) the selection was with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2016 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or AsianAsian for both panels. AgainTypically, note sample allocations across sample domains change from one MEPS panel to another. The sample domains used may also vary by panel although this was not the case for Panel 20 and Panel 21. When one compares unweighted measures (e.g., response rates) between panels and years, one should take into account such differences. If, for example, members of one domain have a lower propensity to respond than those of another domain, then if that not all NHIS domain has been allocated a higher proportion of the sample, the corresponding panel may have a lower unweighted response rate simply because of the differences in sample allocation. Within each domain (sample stratum) systematic samples of the MEPS-eligible households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete from among the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnairehousehold respondents made available for MEPS sample selection purposes.

Oversampling. Xxxxxxxxxxxx was employed for selected subgroups of policy-level interest to help increase the precision of estimates associated with members of those subgroups. Before going into details, the concept of oversampling is discussed. In a sample where all persons in a population are selected with the same probability and survey coverage of the population is high, the sample distribution is expected to be proportionate to the population distribution. For example, if Hispanics represent 15 percent of the general population, one would expect roughly 15 percent of the persons sampled to be Hispanic. However, in order to improve the precision of estimates for subgroups of a population, one might decide to select samples from those subgroups at higher rates than the remainder of the population. Thus, one might select Hispanics at twice the rate (i.e., at double the probability) of persons not oversampled. As a result, subgroups that are “oversampled” are represented at disproportionately high rates in the sample. Sample weights help ensure that population estimates account for this disproportionate contribution from oversampled subgroups, as the base sample weights for oversampled groups will be smaller than for the portion of the population not oversampled. For example, if a subgroup is sampled at roughly twice the rate of sample selection for the remainder of the population not oversampled, members of the oversampled subgroup will receive base or initial sample weights (prior to nonresponse or poststratification adjustments) that are roughly half the size of the group “not oversampled”. As mentioned above, oversampling a subgroup is done to improve the precision of survey estimates for that particular subgroup. The “cost” of oversampling is that the precision of estimates for the general population and subgroups not oversampled will be reduced to some extent compared to the precision one could have achieved if the same overall sample size were selected without any oversampling. For MEPS, some of the oversampling was achieved through its linkage to the NHIS. For the earlier sample design of the NHIS, Hispanic households were oversampled at a rate of roughly 2 to 1. That iswhere Hispanics, the probability of selecting a Hispanic household for participation in the NHIS was roughly twice that for households in the general population that were not Blacks, and Asians are oversampled. The NHIS oversampling rate for black households was roughly 1.5 to 1. For the new NHIS sample design Asians are also oversampled. The estimated overall oversampling rates associated with each of the three minorities have not yet been reported. The oversampling approaches and the sampling domains used approach for MEPS Panels 15 17 and 16 18 used the same sampling domains (strata) for subsampling among the NHIS respondents eligible for MEPS were somewhat differentMEPS. For Panel 15 four Five domains (strata) were established in a hierarchical sequence. The : the first stratum domain contained households with Asians, a ; the second stratum domain contained households with Hispanics not assigned to the first stratum, a domain; the third stratum domain contained households with Black members not assigned to the first two strata, domains one and two; the fourth stratum domain contained all those remaining households (“Other” households. For Panel 16 the corresponding hierarchical ordering of the domains for sampling among NHIS responding households was: households where an individual was identified in the NHIS as having cancer via the NHIS sampled adult questionnaire; Asian; Hispanic; Black; Other households ) characterized as “complete” for the NHIS; and Other households the fifth domain contained all remaining “Other” households, those characterized as “partial complete” for the NHIS. For In terms of sampling rates, for Panel 15, 17 all households in the Asian domain Asian, Hispanic, and Black domains were sampled with certainty. The sampling rate was about 76 51 percent for the Hispanic domain, about 86 percent for the Black “Other complete” domain, and about 61 40 percent for the “OtherOther partial complete” domain. For Panel 16 18 the corresponding sampling rates for all the domains except those characterized as “Other” were all 1 (i.e., all households assigned to those domains were included in MEPS). The “Other, complete” domain was sampled at a rate of about 79 63 percent while the “Other, partial complete” domain was sampled at a rate of about 46 43 percent. Within strata (domains) for both panels, responding NHIS households were selected for MEPS using a systematic sample selection procedure from among those eligible. For all but those strata involving “Others”, the selection was with equal probability. Within the two strata involving “Others” (for both Panel 15, one stratum; for 17 and Panel 16, two strata18) the selection sample was selected with probability proportionate to size (pps) where the size measure was the inverse of the NHIS initial probability of selection. The pps sampling was undertaken to help reduce the variability in the MEPS weights incurred due to the variability of the NHIS sampling rates. With the subsampling, households that were oversampled for MEPS in calendar year 2011 2013 were those responding households in the NHIS identified as having members with cancer or whose race/ethnicity was Hispanic, Black, or Asian. Again, note that not all NHIS households where a member had cancer were identified as such - the member with cancer had to have been randomly selected to complete the NHIS “sampled adult” questionnaire and to have self-identified as having cancer in response to questions from that questionnaire.