|
|
MAKING IT USEABLE
Interpreting Estimates
Sampling Error for Subsets of the NPTS Data
As noted in the preceding segment, sampling errors cannot be estimated precisely for subsets of the data if those subsets are not random subsets.
If a subset of data is chosen on the basis of any characteristic in the data, such as household size, geographic location, vehicle ownership, etc. it is by definition nonrandom, unless the subset comprises one or more entire strata from the stratified sample design. Although not recommended it is possible that by selecting geographic regions that comprise specific strata of the original sample, we can estimate some sampling errors for smaller geographic subsamples.
Table 9
Comparison of Sampling Errors by Geographic Level
Measure
Statistics
Nation
South
Louisiana
New Orleans
Sample Size 42,033
11,121
354
101
Household Size Mean 2.584
2.554
2.606
2.527
Standard
Deviation1.409
1.371
1.308
1.183
Sampling
Error (SE)0.010
0.016
0.071
0.118
SE as
% of Mean0.37%
0.63%
2.71%
4.66%
Vehicles per Household Mean 1.779
1.796
1.746
1.613
Standard
Deviation1.036
0.993
0.967
0.987
Sampling
Error (SE)0.007
0.012
0.051
0.098
SE as
% of Mean0.36%
0.64%
2.91%
6.09%
Work Trips per Household Mean 2.790
2.816
2.628
3.243
Standard
Deviation3.710
3.760
3.628
4.316
Sampling
Error (SE)0.025
0.044
0.192
0.429
SE as
% of Mean0.90%
1.57%
7.31%
13.24%
Workers per Household Mean 1.299
1.282
1.260
1.403
Standard
Deviation0.959
0.939
0.878
0.874
Sampling
Error (SE)0.006
0.011
0.047
0.087
SE as
% of Mean0.50%
0.85%
3.72%
6.20%
Annual Vehicle Miles of Travel per Vehicle Mean 12,226.0
13,108.2
14,308.8
14,003.5
Standard
Deviation12,286.0
12,977.0
16,507.2
14,581.2
Sampling
Error (SE)66.3
120.8
727.2
1,219.2
SE as
% of Mean0.54%
0.92%
5.08%
10.21%
Bicycle Trips per Household Mean 0.082
0.056
0.104
0.086
Standard
Deviation0.635
0.482
0.721
0.546
Sampling
Error (SE)0.004
0.006
0.039
0.054
SE as
% of Mean5.22%
10.13%
37.12%
63.28%
The table above shows that, for most statistics, the means change a little from level to level, largely as a function of geographic differences, while the sampling errors increase with decreasing geography, as one would expect.
