1. The 2008 NYCHVS used the same sample as in 2005 and 2002, drawn from Census 2000 andupdated by including newly created housing units. The 1991 NYCHVS sample was selected from the 1990 census and updated for the 1993, 1996, and 1999 NYCHVSs.

2. The 2008, 2005, and 2002 NYCHVSs were weighted based on the results from Census 2000 , while the 1991, 1993, 1996, and 1999 NYCHVSs were all weighted based on results from the 1990 decennial census. Since surveys based on different censuses produce different results, it may be difficult to compare data from the 2008, 2005, and 2002 NYCHVS to the 1999 NYCHVS and earlier years.

1. The NYCHVS surveys from the 1990s and from the 2000s are from different samples and are weighted based on different decennial censuses (1990 and 2000). We know that Census 2000 had a better coverage of population and housing in the United States and in New York City than did the 1990 census. Therefore it is difficult to compare data from the 2000/2005/2008 NYCHVS with those from 1999 or earlier.
2. In Census 2000, there was a major change in the way race data was collected in the decennial census. Respondents were allowed to choose more than one race, whereas prior to Census 2000, respondents could only select one race. This was an important difference in the Census 2000 from earlier censuses. All current surveys had to make this change in data collection. So, for the 2002/2005/2008 NYCHVS, respondents were allowed to select more than one race, whereas in the 1999 NYCHVS and earlier surveys, only one race could be selected.

In comparing data from 1999 and previous NYCHVSs with data from 2002/2005/2008 NYCHVSs data users should compare percentages, means, and medians, rather than absolute numbers of housing units, households, and/or persons.

Some rental housing units in the survey occur in more than one rent regulation category. In order to assign such units to one mutually exclusive category, units matched to the sample file that were both HUD-regulated [because they receive subsidies from the U.S. Department of Housing and Urban Development (HUD)] and also rent stabilized, were classified in 2002 and previous HVSs in the final Control Status Recode as 'HUD Regulated'.

However, under this classification system fluctuations in the number of rent-stabilized units which were also HUD Regulated units (matched to the sample file) have created fluctuations in estimates of the number of rent-stabilized units. In order to have a consistent estimate of rent-stabilized units, the Census Bureau revised the rent regulation classification system in 2005 and 2008 to classify all rent-stabilized units as rent-stabilized, even if they are also HUD regulated. The revised classification provides better survey-to-survey comparability of rent stabilized units.

The micro-data files for 2008, 2005, and 2002 (re-issued) incorporate two rent regulation classification codes, one under the old system (Control Status Recode) and one under the revised system (New Control Status Recode).

The New York City Housing and Vacancy Survey (NYCHVS) is a sample survey. Data from the NYCHVS are subject to sampling and non-sampling errors. Data users should have an understanding of standard errors of the data, as a measure of statistical reliability, particularly where sample sizes are small.

The NYCHVS is principally designed to provide statistically reliable data for New York City as a whole and for each of the five boroughs. Data for sub-borough areas are generally much less reliable than city-wide and borough data. Comparisons of sub-borough area data between survey years should be done with great caution, since the sample size for housing and household characteristics for sub-borough areas is small, relative to the entire city. The reliability of estimates of changes in such characteristics between surveys is likely to be extremely low, as seen below.

The sample estimate and its standard error enable one to construct a confidence interval. A confidence interval is a measure of an estimate's reliability. The larger the confidence interval in relation to the size of the estimate, the less reliable the estimate. For example, the standard error on the estimated rental vacancy rate of 1.86 in Borough Park in Brooklyn is 1.94 percentage points. The 90-percent confidence interval is calculated as 1.86 percent +/- (1.645 x 1.94), or 1.86 percent +/- 3.19, or from 0 percent to 5.05 percent. This means that if a census of every housing unit in the sub-borough had been taken, using exactly the same procedures as in the NYCHVS, there is a 90 percent chance that the actual rental vacancy rate would fall within the confidence interval.

Another example: In Borough Park, Brooklyn the estimated standard error on the number of renter housing units (35,357) in Borough Park, Brooklyn is 3,030. The 90-percent confidence interval is calculated as 35,357 +/- (1.645 x 3,030), or 35,357 +/- 4,984, or from 30,373 to 40,341.

As a final example - the rental vacancy rate in Stuyvesant Town/Turtle Bay in Manhattan was 2.89 percent in 2005, compared to 3.55 percent in 2008. Thus, the apparent difference between the two rates is 0.66 percent. The standard errors were 0.91 percent and 0.77 percent respectively. This means that the 90-percent confidence interval for the 2005 estimate of 2.89 covered the range 1.39 through 4.39. The 90-percent confidence interval for the 2008 estimate of 3.55 includes the range 2.28 through 4.82. The two confidence intervals overlap from 2.28 to 4.39. Therefore, we can conclude that the rental vacancy rates in Stuyvesant Town/Turtle Bay in 2008 and 2005 are not statistically different from each other at the 90-percent confidence interval.

For more details, the document on sample design, estimation procedures, accuracy of the estimates, and Topcoding is presented in Source and Accuracy of the Estimates. Estimated standard errors for many HVS data items are provided at Standard Errors of the Estimates.