MAKING IT USEABLE    Interpreting Estimates       Limitations of NPTS

NPTS was not designed to provide data for specific metropolitan areas, nor for townships, counties, or states.

• NPTS can be used reasonably for subregions, consisting of several states, but any lower geography is quite suspect.
• These cautions do not apply to those states and urban areas where add-on samples were obtained.
• Sampling errors are shown in Table 11 for several states and for several large urban areas, to show the problems associated with smaller geography.

In the past, there have been many misuses of NPTS data for analysis, modeling, and other potential uses.

• Not matching results to published control totals, producing the risk that an error is made in tabulating data and incorrect conclusions drawn from the analysis.
• Using geographic samples that are too small for the data to support and not reporting the error levels and confidence bounds associated with them.
• Cutting the dataset too fine in other than geographic ways, e.g., transit trips by rural residents on a Tuesday between noon and 4 p.m.

Using unweighted data can lead to a serious misunderstanding of what the data show. Here are some examples.

• Computing trips and travel from the summary number of trips or miles in the household file, instead of developing the estimates from the travel-day file, because the weighting in the household file is not the correct one for estimating total trips and travel.
• Not understanding that the person weights are adjusted to compensate for missing household members, and using household weights when person or travel day weights should be used.
• Not understanding that when a person has zero trips on the travel day, he or she was interviewed for the travel day but did not leave home all day (or refused to tell the interviewer about any travel).
• Not remembering to omit the legitimate skip, refused, and not reported responses, thus doing such things as adding in 999,999 miles for a vehicle for which mileage is actually missing. There are codes for missing values inserted in the data. Usually these are set by filling the available range for the data item with 9s e.g., 999999 for vehicle mileage. If these values are not declared as missing to such analysis programs as SAS, SPSS, or a spreadsheet, then they will be included as legitimate values in computing statistics. For example, forgetting to omit the legitimate skip (999997), refused (999998) and not-reported (999999) values in the vehicle miles of travel data would result in an estimate of about 134,000 miles per vehicle per year, instead of the correct value of 12,226 miles per vehicle per year.

---