Data on kitchen facilities were obtained from Housing Question 10 in the 2006 American Community Survey. The question was asked at both occupied and vacant housing units. A unit has complete kitchen facilities when it has all three of the following facilities: (d) a sink with a faucet, (e) a stove or range, and (f) a refrigerator. All kitchen facilities must be located in the house, apartment, or mobile home, but they need not be in the same room. A housing unit having only a microwave or portable heating equipment such as a hot plate or camping stove should not be considered as having complete kitchen facilities. An icebox is not considered to be a refrigerator.

The 1996-1998 American Community Survey questions asked whether the house or apartment had complete kitchen facilities, requiring that the three facilities all be in the same unit. In 1999, the American Community Survey added "mobile home" to the question, and capitalized the word "COMPLETE" for emphasis.

Since the 2006 data collection year, estimates from the ACS have included data from both people living in both HUs and GQs. The weighting of GQ persons is performed in three major steps. The first step calculates the sampling base weights, which includes adjustments for subsampling that occurs at the time of interview. The second step adjusts the interviewed person records for nonresponse. The third step adjusts the person weights so that the weighted estimates conform to estimates from the PEP at the state by major GQ type group level. The basic weighting area used for the GQ weighting is the state.

Since 1996, the American Community Survey questions have been the same. Starting in 2004, 'meals included in rent' is shown for all renter-occupied housing units. In previous years (1996-2003), it was shown only for specified renter-occupied housing units.

A noninterview adjustment factor is calculated to account for the eligible GQ residents who do not complete an interview. This occurs in a single step where the noninterview adjustment cells are defined, within state, by major GQ type group by county. If a cell contains fewer than 10 interviews and has any number of noninterviews or if the noninterview factor is greater than 2, then cells are collapsed across counties within the same major GQ type group in an attempt to preserve the state by type group weighted totals. If the new collapsed cell still fails one or both of the collapsing criteria, then it is collapsed to a subset of the type groups within the same institutional/ noninstitutional class as shown in Table 11.2. If needed, all cells with the same institutional/ noninstitutional class are collapsed together across all type groups in the class. If further collapsing is still required, then all cells within the state are collapsed together. In practice, these last two collapsings are rarely, if ever, used. The GQ Noninterview Adjustment Factor ( GQNIF ) for each eligible cell is then calculated:


All interviewed GQ persons are adjusted by this noninterview factor. All noninterviews including those persons who were found to be out-of-scope are assigned a factor of 0.0. The computation of the weight after the noninterview adjustment factor is summarized in Table 11.3.

Table 11.3 Computation of the Weight After the GQ Noninterview Adjustment Factor ( WGQNIF )

The 1996-1998 American Community Survey questions were the same. Starting in 1999, the question had a lead-in question on whether the respondent had an installment loan or a contract on the mobile home. The question then asked for total costs including any installment loan.

Single-year weighting is implemented in three stages. In the first stage, weights are computed to account for differential selection probabilities based on the sampling rates used to select the HU sample. In the second stage, weights of responding HUs are adjusted to account for nonresponding HUs. In the third stage, weights are controlled so that the weighted estimates of HUs and persons by age, sex, race, and Hispanic origin conform to estimates from the PEP of the Census Bureau at a specific point in time. The estimation methodology is implemented by "weighting area," either a county or a group of less populous counties.

This measure divides the monthly housing costs distribution into two equal parts: one-half of the cases falling below the median monthly housing costs and one-half above the median. Medians are shown separately for units "with a mortgage" and for units "not mortgaged." Median monthly housing costs are computed on the basis of a standard distribution. (See the "Standard Distributions" section under Derived Measures.") Median monthly housing costs are rounded to the nearest whole dollar.

The first step is to compute the basic sampling weight for the HU based on the inverse of the probability of selection. This sampling weight is computed as a multiplication of the base weight ( BW ) and a computer-assisted personal interviewing (CAPI) subsampling factor ( SSF ). The BW for an HU is calculated as the inverse of the final overall first-phase sampling rate as given in Chapter 4, Table 4.2. HUs sent to CAPI are eligible to be subsampled (second-phase sampling) at one of the rates described in Table 4.4. Those selected for the CAPI subsample, and for which no late mail return is received in the CAPI month, are assigned a CAPI SSF equal to the inverse of their (second-phase) subsampling rate. Those not selected for the CAPI subsample receive a factor of 0.0. HUs for which a completed mail return is received, regardless if it was eligible for CAPI, or a computer-assisted telephone interviewing (CATI) interview is completed receive a CAPI SSF of 1.0. The CAPI SSF is then used to calculate a new weight for every HU, the weight after CAPI subsampling factor ( WSSF ). It is equal to the base weight times the CAPI subsampling factor. After each of the subsequent weighting steps, with one exception that will be noted, a new weight is calculated as the product of the new factor and the weight following the previous step. For additional details about the weighting steps discussed in this and the following section, see Asiala (2004).

Table 11.4 Computation of the Weight After CAPI Subsampling Factor ( WSSF )

Weighting step	Sample disposition
	Mail respondent	CATI respondendent	CAPI sampled units	CAPI nonsampled units	CAPI eligible, but then becomes mail respondent
Base Weight (BW)	1 ÷ (overall sampling rate)	1 ÷ overall sampling rate)	1 ÷ (overall sampling rate)	1 ÷ (overall sampling rate)	1 ÷ (overall sampling rate)
CAPI subsampling factor (SSF)	1	1	1 ÷ (CAPI sub- sampling rate)	0	1
Weight after subsampling factor (WSSF)= BW × SSF	1 ÷ (overall sampling rate)	1 ÷ (overall sampling rate)	1 ÷ (overall sampling rate) × 1 ÷ (CAPI sub- sampling rate)	0	1 ÷ (overall sampling rate)

Note: Table summarizes computation of the WSSF by the weighting step and the sample dispostion.

The goal of ACS estimation is to represent the characteristics of a geographic area across the specified period. For single-year estimates, this period is 12 months, and for 3- and 5-year estimates, it is 36 and 60 months, respectively. The annual sample is allocated into 12 monthly samples. The monthly sample becomes a basis for the operations of the ACS data collection, preparation, and processing, including weighting and estimation.

The data for HUs assigned to any sample month can be collected at any time during a 3-month period. For example, the households in the January sample month can have their data collected in January, February, or March. Each HU in a sample belongs to a tabulation month (the month the interview is completed). This is either the month the processing center checked in the completed mail questionnaire or the month the interview is completed by CATI or CAPI.

Because of seasonal variations in response patterns, the number of HUs in tabulation months may vary, thereby over-representing some months and under-representing other months in the single and multiyear estimates. For this reason, an even distribution of HU weights by month is desirable. To smooth out the total weight for all sample months, a variation in monthly response factor ( VMS ) is calculated for each month as:



This adjustment factor is computed within each of the 2,005 ACS weighting areas (either a county or a group of less populous counties). The index for weighting area is suppressed in this and all other formulas for weighting adjustment factors.

Table 11.5 illustrates the computation of the VMS adjustment factor within a particular county. In this example, the total base weight ( BW ) for each month is 100 (as shown on line 1 of this table). The total weight ( WSSF ) across modes within each month varies from 90 to 115 (as shown on line 5). The VMS factors are then computed by month as the ratio of the total BW to the total WSSF (as shown on line 6).

Table 11.5 Example of Computation of VMS

Line	Month
	March	April	May	June	July
Line 1: Total base weight (BW) across released samples Total weight after CAPI subsampling (WSSF)by mode:	100	100	100	100	100
Line 2: (a) Mail	55 (Mar sample)	45 (Apr sample)	40 (May sample)	45 (Jun sample)	50 (Jul sample)
Line 3: (b) CATI	30 (Feb sample)	25 (Mar sample)	30 (Apr sample)	30 (May sample)	25 (Jun sample)
Line 4: (c) CAPI	30 (Jan sample)	25 (Feb sample)	20 (Mar sample)	25 (Apr sample)	30 (May Sample)
Line 5: Total weight WSSF across modes (a+b+c)	115	95	90	100	105
Line 6: VMS adjustment factor	100 ÷ 115	100 ÷ 95	100 ÷ 90	100 ÷ 100	100 ÷ 105

The adjusted weights after the variation of monthly response adjustment ( WVMS ) are a product of the weights after CAPI subsampling factor ( WSSF ) and the variation of monthly response factor ( VMS ). When the VMS factor is applied, the total VMS weights ( WVMS ) across all HUs tabulated in a sample month will be equal to the total base weight of all HUs selected in that months sample. The result is that each month contributes approximately 1/12 to the total single-year estimates. In other words, the single-year estimates of ACS characteristics are a 12-month average without over- or under-representing any single month due to variation in monthly response. Analogously, each month contributes approximately 1/36 and 1/60 to the 3- and 5-year estimates, respectively.

Since 1996, the American Community Survey questions have been the same.

In this step, all HUs are placed into adjustment cells based on the cross-classification of building type (single- versus multiunit structures) and census tract. If a cell contains fewer than 10 interviewed HUs, it is collapsed with an adjoining tract until the collapsed cell meets the minimum size of 10.² Cells with no noninterviews are not collapsed, regardless of size, unless they are forced to collapse with a neighboring cell that fails the size criterion. The first noninterview adjustment factor ( NIF1 ) for each eligible cell is:


Footnote:

²Data are sorted by the weighting area, building type, and tract. Within a building type, a tract that has 10 or more responses is put in its own tract. A tract that has no nonresponses and some responses (even though the total is fewer than 10) is put in its own tract. A tract that has nonresponses and fewer than 10 responses is collapsed with the next tract. If the final tract needs to be collapsed, it is collapsed with the previous tract.

where
WVMS ij = Adjusted HU weight after the variation in monthly response adjustment for the j th HU within the i th adjustment cell.

All occupied and temporarily occupied interviewed HUs are adjusted by this first noninterview factor. Vacant and deleted HUs are assigned a factor of 1.0, and noninterviews are assigned a factor of 0.0. The computation of the weight after the first noninterview adjustment factor is summarized in Table 11.6.

Table 11.6 Computation of the Weight After the First
Noninterview Adjustment Factor ( WNIF1 )

Interview status	WNIF1ij
Occupied or temporarily occupied HU	WVMSij × NIF1i
Vacant or deleted HU	WVMSij
Noninterviewed HU	0

Since 1996, the American Community Survey questions have been the same.

One element not accounted for by the two noninterview factors above is the systematic differences that exist between characteristics of households that return census mail forms and those that do not (Weidman et al., 1995). The same element has been observed in the ACS across response modes. Virtually all noninterviews occur among the CAPI sample, and people in these HUs may have characteristics that are more similar to CAPI respondents than to mail and CATI respondents. Since the noninterview factors ( NIF1 and NIF2 ) are applied to all HUs interviewed by any mode, compensation may be needed for possible mode-related noninterview bias. The mode bias factor ensures that the total weights in the cells defined by a cross-classification of selected characteristics are the same as if the weight of noninterview HUs had been assigned only to CAPI HUs, but the factor distributes the weight across all respondents (within the cells) to reduce the effect on the variance of the resulting estimates.

The first step in the calculation of the mode bias noninterview factor ( MBF ) is to calculate an intermediate factor, referred to as the mode noninterview factor ( NIFM ). The NIFM is not used directly to compute an adjusted weight; instead, it is used as a factor applied to the WVMS weight to allow the calculation of the MBF . The cross-classification cells are defined for building type by tabulation month. Only HUs interviewed by CAPI and noninterviews are placed in the cells. If a cell contains fewer than 10 interviewed HUs, it is collapsed with an adjoining month. Cells with no noninterviews are never collapsed unless they are forced to collapse with a neighboring cell that fails the size criterion. The NIFM for a cell is:



This mode noninterview factor is assigned to all CAPI-interviewed occupied and temporarily occupied HUs. HUs for which interviews are completed by mail or CATI, vacant HUs, and deleted HUs are given a factor of 1.0. Noninterviews are given a factor of 0.0. The NIFM factor is used in the next step only. Note that the NIFM adjustment is applied to the WVMS weight rather than the HU weight after the first and second noninterview adjustments ( WNIF1 and WNIF2 ). The computation of the weight after the mode noninterview adjustment factor is summarized in Table 11.8.

Table 11.8 Computation of the Weight After the Mode
Noninterview Adjustment Factor ( WNIFM )

Interview status	WNIFMij
Occupied or temporarily occupied HU	WVMS 1 ij × NIFMi
Vacant or deleted HU	WVMSij
Noninterviewed HU	0

where
WNIFMij = Adjusted HU weight after the mode noninterview adjustment for the j th HU within the i th adjustment cell.

Next, a cross-classification table is defined for tenure (three categories: HU owned, rented, or temporarily occupied), tabulation month (12 categories), and marital status of the householder (three categories: married/widowed, single, or unit is temporarily occupied). All occupied and temporarily occupied interviewed HUs are placed in their cells. If a cell has fewer than 10 interviewed HUs, the cells with the same tenure and month are collapsed across all marital statuses. If there are still fewer than 10 interviewed HUs, the cells with the same tenure are collapsed across all months. The mode bias factor ( MBF ) for each cell is then calculated as:



All interviewed occupied and temporarily occupied HUs are adjusted by this mode bias factor, and the remaining HUs receive the factor 1.0. These adjustments are applied to the WNIF2 weights. The computation of the weight after the mode bias factor is summarized in Table 11.9 below.

Table 11.9 Computation of the Weight After the Mode
Bias Factor ( WMBF )

Interview status	WMBFij
Occupied or temporarily occupied HU	WNIF 2ij x MBFi
Vacant, deleted or noninterviewed HU	WNIF 2ij

where
WMBFij = Adjusted HU weight after the mode bias factor adjustment for the j th HU within the i th adjustment cell.

This stage of weighting forces the ACS total HU and person weights to conform to estimates from the Census Bureau's PEP. The PEP of the Census Bureau annually produces estimates of population by sex, age, race, and Hispanic origin, and total HUs for each county in the United States as of July 1. The ACS estimates are based on a probability sample, and will vary from their true population values due to sampling and nonsampling error (see Chapters 12 and 14). In addition, we can see from the formulas for the adjustment factors in the previous two sections that the ACS estimates also will vary based on the combination of interviewed and noninterviewed HUs in each tabulation month. As part of the process of calculating person weights for the ACS, estimates of totals by sex, age, race, and Hispanic origin are controlled to be equal to population estimates by weighting area. There are two reasons for this: (1) to reduce the variability of the ACS HU and person estimates, and (2) to reduce bias due to under-coverage of HUs and the people within them in household surveys. The bias that results from missing these HUs and people is partly corrected by using these controls (Alexander et al., 1997).

The assignment of final weights involves the calculation of three factors based on the HU and population controls. The first adjustment involves the independent HU estimates. A second and separate adjustment relies on the independent population estimates. The final adjustment is implemented to achieve consistency between the ACS estimates of occupied HUs and householders.

The Census Bureau produces estimates of total HUs for states and counties as of July 1 on an annual basis. The estimates are computed based on a model:

HU0X = HU00 + (NC0X + NM0X) − HL0X

where the suffix "X" indicates the year of the HU estimates, and

HU0X = Estimated 200X HUs

HU00 = Geographically updated Census 2000 HUs

NC0X = Estimated residential construction, April 1, 2000, to July 1, 200X

NM0X = Estimated new residential mobile home placements, April 1, 2000, to July 1, 200X

HL0X = Estimated residential housing loss, April 1, 2000, to July 1, 200X.

For more detailed background on the current methodology used for the HU estimates, readers can visit and select "Housing Unit Estimates."

The Census Bureau also produces population estimates as of July 1 on an annual basis. Those estimates are computed based on the following simplified model:

P1 = P0 + B − D + NDM + NIM + NMM ,

where

P1 = population at the end of the period (current estimate year)
P0 = population at the beginning of the period (previous estimate year)
B = births during the period
D = deaths during the period
NDM = net domestic migration during the period
NIM = net international migration during the period
NMM = net military movement during the period.

In practice, the model is considerably more complex to leverage the best information available from multiple sources. For more detailed background on the current methodology used for the population estimates, readers can visit and select "State and County Population Estimates."

Production of the population estimates for Puerto Rico is limited to population totals by municipio , and by sex-age distribution at the island level. For this reason, estimates of totals by municipio , sex, and age for the PRCS are controlled so as to be equal to the population estimates. Currently, there are no HU controls available for Puerto Rico.

Note that both HU and population estimates used as controls have a reference date of July 1 which means that the 12-month average of ACS characteristics is controlled to the population with the reference date of July 1. If person weights are controlled to the population estimates as of that date, it is logical that HUs also are controlled to those estimates to achieve a consistent relationship between the two totals.

The HU post-stratification factor is employed to adjust the estimated number of ACS HUs by weighting area to agree with the PEP estimates. For the i th weighting area, this factor ( HPF ) is:

HPFi = PEP HU estimate

÷

Total adjusted HU weight after the mode bias factor of interviewed occupied, interviewed temporarily
occupied, and vacant HUs



where
HUi = PEP housing unit estimate for the i th weighting area.

The denominator of the HPF formula aggregates the adjusted HU weight after the mode bias factor adjustment ( WMBF ) across 12 months for the interviewed occupied, interviewed temporarily occupied, and vacant HUs. All HUs then are adjusted by this HU post-stratification factor. Therefore, WHPF = WMBF × HPF , where WHPF is the adjusted HU weight after the HU post-stratification factor adjustment.

The 1996-2006 American Community Survey questions were stand-alone questions that asked the respondent to answer either Yes, has all three facilities or No to the question of whether the housing unit had complete plumbing facilities, requiring that the facilities all be in the same unit. Starting in 2006, the structure of the question changed and combined plumbing facilities with kitchen facilities and telephone service availability into one question to ask, Does this house, apartment, or mobile home have - and provided the respondent with a Yes or No checkbox for each component needed for complete facilities. An additional change introduced In 2006 included changing the description of the component hot and cold piped water to hot and cold running water.

Prior to the calculation of person weights, each HU has a single weight which is independent of the characteristics of the persons residing in the HU. After the calculation of person weights, a new HU weight is computed by taking into account the characteristics of the householder in the HU. In each interviewed occupied HU, the householder defined as the reference person (one of the persons who rents or owns the HU) is identified. Adjustment of the HU weight to account for the householder characteristics is done by assigning a householder factor ( HHF ) for an HU equal to the person post-stratification factor ( PPSF ) of the householder.⁴ Their PPSF s give an indication of under-coverage for households whose householders have the same demographic characteristics. The HHF adjustment uses this information to adjust for the resultant bias. Vacant HUs are given an HHF of 1.0 because they have no householders.

The adjusted HU weight accounting for householder characteristics is computed as a multiplication of the adjusted HU weight after the HU post-stratification factor adjustment ( WHPF ) with the householder factor ( HHF ). Therefore, WHHF = WHPF × HHF , where WHHF is the adjusted HU weight after the householder factor adjustment. The HU weight after the householder factor adjustment becomes the final HU weight.

The ACS weighting procedure results in two separate sets of weights, one for HUs and one for persons residing within HUs. However, since the housing unit weight is equal to the person weight of the householder, the survey will produce logically consistent estimates of occupied housing units, households, and householders. With this weighting procedure, the survey estimate of total housing units will differ slightly from the PEP total housing unit estimates. The difference between the ACS estimate the PEP estimate nationally, however, was less than 5,000 in 2006.

Footnote:

⁴In the calculation of person weights, the PPSF is used to adjust person weight so that the ACS population estimates conform to PEP estimates by demographic characteristics.

The multiyear estimation methodology involves reweighting the data for each sample address in the 3- or 5-year period and is not just a simple average of the single-year estimates. The weighting methodology for the multiyear estimation is very similar to the methodology used for the single-year weighting. Thus, only the differences between the single- and multiyear weighting are described in this section.

The data for all sample addresses over the multiyear period are pooled together into one file. The single-year base weights are then adjusted by the reciprocal of the number of years in the period so that each year contributes its proportional share to the multiyear estimates. For example, for the 2005−2007 3-year weighting, the base weights are all divided by three.

The interview month assigned to each address is also recoded so that all the data from the entire period appears as though it came from a 1-year period. For example, in the 2005−2007 3-year weighting, all addresses that were originally assigned an interview month of January 2005, 2006, or 2007 are assigned the common interview month of January. Thus, when the weighting is performed, those records will all be treated as though they come from the same month for the VMS , NIF2 , NIFM , and MBF adjustments. By pooling the records across years in this manner, the noninterview adjustments, in particular, require less collapsing because of the larger sample in each cell. This, in turn, should better preserve the seasonal trends that may be present in the population as captured by the ACS.

Since 1996, the American Community Survey questions have been the same.

Since the multiyear estimate is an estimate for the period, the controls are not those of a particular year but rather they are the average of the annual independent population estimates over the period. The PEP refreshes their entire time series of estimates going back to the previous census each year using the most current data and methodology. Each of these time series are considered a "vintage." In order for the ACS to make use of the best available population estimates as controls, the multiyear weighting uses the population estimates of the most recent vintage for all years in the period in order to derive the multiyear controls.

These derived estimates are created for the HU, GQs population, and total population for use as controls in the multiyear weighting. The derived county-level HU estimates are the simple average across all years in the period. Since the average is typically not an integer, the result is rounded to the final integerized estimate. Likewise, the derived GQ population estimates for state by major type group are the simple average across all years in the period. Those averages are then control rounded so that the rounded state average estimate is within one of the unrounded estimate. Finally, the derived total population estimates by race, ethnicity, age, and sex are averaged across all years in the period and control rounded to form the final derived estimates. This is done prior to the collapsing of the estimates into the 156 cells per weighting area needed for the demographic dimension of the household person weighting as described in the single-year person weighting section.

Once the data are pooled and put into the geography of the final year, they are weighted using the single-year weighting methodology through the MBF adjustment. It is after this adjustment that the only weighting step specific to the multiyear weighting methodology is implemented, the model-assisted estimation procedure. An earlier research project (Starsinic, 2005) compared the variances of ACS tract-level estimates formed from the 1999−2001 ACS to the variances of the Census 2000 long-form estimates. The results of that research showed that the variances of the ACS tract-level estimates were higher in relation to the long form than what we expected based on sample size alone. The primary source of that increased variance was attributed to the lack of ACS subcounty controls at the tract-level or lower as was used for the long form.

Several options were explored on how the ACS estimates of variance for subcounty estimates might be improved. One option considered was to use the ACS sampling frame counts as subcounty controls. Other options explored ways to create subcounty population controls, including tract-level population controls. The final approach, and the one that was chosen, introduces a model-assisted estimation step into the multiyear weighting that makes use of both the sampling frame counts and administrative records to reduce the level of variance in the subcounty estimates (Fay, 2006). An important feature of the model-assisted estimation procedure is that the administrative record data is not used directly to produce ACS estimates. The administrative record data are only used to help reduce the level of variance. The published ACS estimates are still formed from weighted totals of the ACS survey data.

The entire model-assisted estimation process is summarized in these steps:

1. Create frame counts for places and Minor Civil Divisions (MCDs) that contain at least 10,000 in population and at least 300 HU addresses (the 5-year estimation will use tracts simply satisfying the latter criterion).

2. Link the administrative records to the ACS sampling frame (the Master Address File [MAF]) and drop administrative records that cannot be linked.

3. Form unweighted place- and MCD-level totals (tract-level for the 5-year estimates) of the linked administrative record characteristics.

4. Apply the WMBF weights at the HU level to the linked administrative records that fall into the ACS sample. The weighted estimates at this step represent (essentially) unbiased estimates of the unweighted totals in Step 2.

5. Using generalized regression estimation, fit a model to calibrate the ACS weights so that the weighted totals from the linked ACS records match the unweighted totals from Step 2 and so that the weighted ACS estimate of HUs match the frame totals in Step 1. The categories of the variables considered in the regression are collapsed or removed as necessary to fit a good model.

6. Proceed with the remaining steps of the ACS weighting starting with the Housing Unit Poststratification (HPF) Factor adjustments, including the person weighting using the derived multiyear controls as described in the preceding section.

The BWs , which reflect the sampling probabilities of selection, should sum to the count of records on the sampling frame at the county and, generally, the subcounty level. However, after the noninterview adjustments the weighted subcounty distribution of the interviewed sample cases can deviate from the original frame distribution. This can impact both the subcounty estimates and the variances on those estimates. The use of frame counts as subcounty controls reestablishes the original distribution of HU addresses on the frame in the weighted sample. For the 3-year weighting, these frame counts are calculated at the place- or MCD-level. If the place or MCD has a PEP population estimate of 10,000 or more then the ACS weights are controlled to those frame counts at that subcounty level. For the 5-year weighting, these frame counts will be computed for tracts. This control to the frame counts is the simplest model and is used if a model with administrative record data cannot be estimated. Otherwise, it is one part of the entire calibration performed in this step.

The administrative record data used for this step is created from linking two primary files maintained by the Data Integration Division at the Census Bureau. The first file includes person characteristics and has been created from a combination of social security and census information. The second file uses administrative records to identify all possible addresses of the persons on the first file. A merged file is then created which contains only the age, sex, race, and Hispanic origin of each person and an identifier that links that person to the best address available in the MAF via a Master Address File ID (MAFID). No other characteristics or publicly identifiable information are present on the file. This file is updated annually to account for new births, death information, and for updated address information.

This measure divides the room distribution into two equal parts: one-half of the cases falling below the median number of rooms and one-half above the median. In computing median rooms, the whole number is used as the midpoint of the interval; thus, the category "3 rooms" is treated as an interval ranging from 2.5 to 3.5 rooms. 'Median rooms' is rounded to the nearest tenth. (For more information on medians, see the discussion under "Derived Measures.")

'Aggregate rooms' is calculated by adding all of the rooms for housing units in an area. A value of "10" is assigned to rooms for units falling within the terminal category, "9 or more rooms." (For more information on aggregates, see "Derived Measures.")

Since 1999, the American Community Survey provided response categories from "1 room" to "9 or more rooms." The 1996-1998 American Community Survey question provided a space for a write-in entry.

In addition to the adjustments to the single-year weighting methodology for weighting the multiyear data, there are other steps involved in the multiyear estimation that are not weighting related. These include standardizing definitions of variables, updating the geography for place of work and migration characteristics, and the adjustment of income, value, and other dollar amounts for inflation over the period. The details of these adjustments are given in Chapter 10.

Since 1996, the American Community Survey questions remained the same.

A housing unit may be a house, an apartment, a mobile home, a group of rooms or a single room that is occupied (or, if vacant, intended for occupancy) as separate living quarters. Separate living quarters are those in which the occupants live separately from any other individuals in the building and which have direct access from outside the building or through a common hall. For vacant units, the criteria of separateness and direct access are applied to the intended occupants whenever possible. If that information cannot be obtained, the criteria are applied to the previous occupants.
Both occupied and vacant housing units are included in the housing unit inventory. Boats, recreational vehicles (RVs), vans, tents, railroad cars, and the like are included only if they are occupied as someone's current place of residence. Vacant mobile homes are included provided they are intended for occupancy on the site where they stand. Vacant mobile homes on dealers' sales lots, at the factory, or in storage yards are excluded from the housing inventory. Also excluded from the housing inventory are quarters being used entirely for nonresidential purposes, such as a store or an office, or quarters used for the storage of business supplies or inventory, machinery, or agricultural products.