SWOV homepage
Nederlands  

Quality aspects

The sample is extrapolated to the population size. This means that the totals or subtotals per mode of transport or province have a certain error margin. We will deal with a number of these aspects in the following paragraphs.

Reweighting and extrapolation

The sample is drawn so that all kinds of groups are proportionately represented.

The response of all these groups was not equal, meaning that the results had to be weighted again.

Reweighting is compensating for the under- and over-representation of certain groups, e.g. degree of urbanisation, age, and car possession by registration year. The journeys are also reweighted by day of the year.

Extrapolation is converting the sample results into population results.

Since 1995, a new weigthing model is used for the NTS weighting and extrapolation, and has been applied to the data from 1985 onwards. The reason for the new model was the fact that the car use trends of the NTS and the CBS Car Panel were getting farther apart from each other.

From 1999 onwards, the telephone accessibility of a household is taken into account. The response chance of a telephone accessible household is about twice as great as with a household not accessible by telephone.

The correction for the trend break also led to a revised reweigthing and extrapolation of the numbers, and has been applied to the data from 1985 onwards.

There is data about households, individuals, and journeys, each with their own extrapolation factor.

In the reweighting and extrapolation process, broader groups were used than those ultimately available in the available data. Take, for example, the case of age groups. The sample number of 18-30 year olds has been reweighted and extrapolated using the population numbers; no account has been made for an over- or under-representation of 18-19 year olds within this group.

Correcting for trend break

In mid 2002, the corrected NTS data for the 1985-1998 period became available.

When the 1999 and 2000 data became available, it became clear that differences had occurred as a result of a number of factors:

The central idea is that the new method produced a better description of reality, and that the old data should be corrected.

The 1985-1998 data were examined at the micro level and adapted according the new coding instructions. The journeys no longer included were those of children playing outdoors and those of lorries. Multiple journeys are recoded to one 'there and one back' journey as done with journeys having the same departure and arrival address. Missing data was calculated using the same methods as the new NTS data processing.

CBS determined the response chance of each person and altered the weighting factors accordingly by using a number of features (Sex * Age group * High/Low educated * Number of main transport modes used + Day of the week). See Swinkels & Konen (CBS, 2002) for a detailed explanation.

 

The corrected 1985-1998 results are structurally lower than the original ones. The distances travelled decreased by an average of 13%. These differences were, of course, larger and smaller when subdivided by age, transport mode, and sex.

 

A larger than average decrease in kilometres travelled (i.e. increase in crash rate) exists for:

A smaller than average decrease or even increase in kilometres travelled (i.e. decrease in crash rate) exists for:

The differences between them are, however, small and have practically no effect on the prioritizing. The ranking of the various groups remains practically unchanged. The groups with the highest crash rates are still the mopedists and slopedists, and the elderly.

Points to keep in mind

There are several points to keep in mind when using NTS data.

 

Special attention should be paid to the definitions used in this survey. Examples are: exceptions, reweighting, extrapolating, 0-11 year olds, and holiday journeys. They often explain differences between publications.

 

As a result of the small samples, especially up to 1994, the margins of some variables are quite large. This applies to:

The NTS does not contain any information about journeys by road types or urban/rural.

Reliability, accuracy, and completeness

Many mistakes can be made when filling in a journey diary. Journeys are forgotten (too short, unimportant, nobody else may know, before-and-after transport in public transport journeys), the distance travelled is estimated incorrectly, the times of day are rounded off, etc.

The 1999 results of the new NTS showed that the number of respondents who did not make any journeys was larger than expected. These people (known as 'zero travellers') apparently had been difficult to reach, and for many of them, this was probably the reason not to reply to the old NTS.

In the old NTS the respondent's estimated distances were coded without being controlled to see whether these estimates were plausible considering the journey time. SWOV's assessment study in 1978 showed that, depending on the transport mode, respondents overestimate distances. In the case of car journeys the estimations were 8% and those by bicycle even 20% too far. These corrections were used to recaculate distances travelled.

Since 1999, the new NTS chose not to correct the estimates anymore. Instead, the check is made on whether the relation between the recorded distance and journey time is acceptable. A maximum speed per mode of transport is used for this. On this basis also the trend break for distances was corrected in the old NTS.

 

In addition to this systematic bias, there are also sampling errors because information is only obtained from a fraction of the population. The size of the error is determined by the sum of two types of error: the bias and the variance. The size of the bias can be determined by a non-response study or an improved observation technique. The variance measures the dispersion around an average value, and this can be estimated from the sample data. Real variance calculations are complicated, and in the next paragraph you will find calculations for a relevant table.

Variance and margins

Even after the considerable increase in sample size for the last few years, the data can have random fluctuations that really have to be kept in mind when interpreting the results. This effect of course increases as subdivisions lead to smaller cells (sample number).

 

In the table below, the size and margin class for important road safety research variables are given.

 

Transp.mode

Age

Pedestrian

Bicycle

Moped

Sloped

Motor/ scooter

Car (driver)

Car (passenger)

Bus/Tram/ Metro/Train

Others

Total

0 – 5

231

480

2

1

0

0

6636

182

28

7560

6 – 11

239

817

2

4

8

0

5249

367

25

6711

12 – 14

82

1296

0

1

1

0

1832

588

23

3823

15 – 17

94

1132

10

270

26

0

1820

1476

45

4873

18 – 19

73

505

7

91

8

1000

1307

2594

28

5613

20 – 24

162

764

11

59

193

4706

3097

4168

75

13235

25 – 29

219

786

2

59

161

8765

3294

2468

103

15857

30 – 34 

249

915

2

52

150

12501

3537

1598

83

19088

35 – 39

259

1091

33

50

109

12863

3179

1840

99

19524

40 – 44

247

1120

11

56

202

11469

2955

1357

99

17517

45 – 49

250

1016

18

33

233

10669

2701

1478

86

16484

50 – 54

236

1077

12

20

67

9428

3042

1200

106

15188

55 – 59

215

872

16

15

39

7746

2934

923

101

12863

60 – 64

212

739

16

10

4

4654

1946

748

29

8358

65 – 69

179

525

5

7

11

2694

1432

616

37

5506

70 – 74

127

379

1

2

5

1720

1189

398

51

3871

75 – 79

91

200

6

6

1

1052

639

280

7

2281

80 +

59

93

7

2

4

478

628

234

6

1511

Total

3226

13807

161

738

1222

89745

47417

22515

1029

179861

Table 1. Kilometres NTS 2003 by age group and mode of transport

Margin <10%

Margin < 20%

Margin ≥ 20%

 

 

 

The size of the sample margin percentages of each of the above cells is given in the table below.

 

Transp. mode
Age

Pedestrian

Bicycle

Moped

Sloped

Motor/ scooter

Car (driver)

Car (passenger)

Bus/Tram/ Metro/Train

Others

Total

 

0 – 5

10.6

9.8

196

176

 

 

10.1

40.5

71.7

9.03

 

6 – 11

7.8

7.4

182

93.9

150

 

10.0

35.9

81.6

8.24

 

12 – 14

14.8

6.6

 

179

196

 

12.8

24.2

86.9

7.89

 

15 – 17

15.0

7.2

109

18.6

75.1

 

14.3

16.9

84.4

8.26

 

18 – 19

12.8

11.9

93.2

30.9

126

19.3

16.4

15.9

73.8

9.94

 

20 – 24

11.3

10.0

146

37.1

145

8.7

14.2

12.4

144

7.13

 

25 – 29

11.2

9.9

165

56.2

73.2

7.5

13.4

17.4

80.1

6.32

 

30 – 34

10.3

10.4

174

40.3

42.5

6.3

11.5

21.2

81.8

5.54

 

35 – 39

8.2

8.1

51.1

64.7

44.4

5.8

11.9

16.2

63.9

5.06

 

40 – 44

9.7

7.8

93.7

62.6

39.8

5.9

11.9

16.9

80.0

4.97

 

45 – 49

8.7

7.8

68.8

54.6

55.8

5.9

12.2

17.5

47.1

5.15

 

50 – 54

9.0

8.5

61.8

52.6

75.8

7.3

11.9

18.6

62.8

5.79

 

55 – 59

9.7

9.0

60.4

47.4

77.3

6.9

11.6

20.9

87.5

5.85

 

60 – 64

10.6

10.7

132

67.7

114

8.6

12.7

24.8

62.6

6.83

 

65 – 69

10.9

11.3

69.9

98.8

147

11.0

14.1

30.6

61.1

8.33

 

70 – 74

10.9

13.8

196

118

78.5

14.0

16.7

33.3

88.2

9.77

 

75 – 79

13.1

15.3

114

94.8

95.8

16.9

18.4

38.9

45.0

11.49

 

80 +

16.8

21.7

120

196

52.0

23.8

26.8

61.3

40.0

17.82

 

Total

2.80

2.48

25.92

12.53

29.17

2.03

3.70

5.24

23.25

1.790

 

Table 2. Relative confidence margins (95%) for the NTS sample by age group and transport mode.

 

Margin10%

Margin 20%

Margin ≥ 20%

 

 

 

The 95% confidence limits mean that the real mobility/exposure lies between the value of Table 1 ± the margin in Table 2. With the real mobility/exposure we mean here the value that would have been obtained if everybody had been surveying during the whole year. Bias errors (overestimating distances, omitting journeys, non-response) would still exist.

For example: the 25-29 year old motorists annually travel 8,765 million kilometres ± 7.5%; i.e. between 8,111 and 9,419 million kilometres.

The margins of the numbers in the column and row totals are considerably smaller. This is logical because the number of individuals/households to calculate this is much larger.

 

When the mobility/exposure is used to estimate the crash rate, we have to use the relative margins of both the number of casualties (≈1.94/√N) and kilometres (table 2) to compare subgroups.

 


Deaths per billion kilometres travelled by transport mode, summed for all ages, including margin of numbers of deaths and the NTS sample margin.

In the figure above, the death rates for various transport modes are given. The death rate of slopedists is lower than that of mopedists, but if the margins are included in the comparison, there is probably the other way round (i.e. mopedists have a larger crash rate than slopedists). This calculation does not include the bias errors. The margin shown must therefore be regarded as a lower limit; in reality the margins are even larger. This figure also illustrates that a subdivision of the death rate by age group for motorized two-wheelers is a risky business.

 

The sample size, i.e. the number of journeys on which the mobility/exposure number is based, is a measure for the margin to be applied to the calculated data.

 

Transp.mode Age

Pedestrian

Bicycle

Moped

Sloped

Motor/ scooter

Car (driver)

Car (passenger)

Bus/Tram/ Metro/Train

Others

Total

0 - 5 

3658

3208

3

2

0

0

7109

131

117

14228

6 - 11 

5926

6682

3

18

4

0

6348

261

36

19278

12 - 14 

1536

5114

0

3

1

0

1701

374

34

8763

15 - 17 

1580

4475

13

540

34

0

1485

955

28

9110

18 - 19 

1234

1734

21

176

12

806

869

1139

18

6009

20 - 24 

2244

2581

14

107

49

3996

1625

1841

19

12476

25 - 29 

2663

2651

4

77

60

6416

1876

910

45

14702

30 - 34 

3778

3895

4

87

91

9808

2160

781

52

20656

35 - 39 

4594

5520

48

72

70

11547

2244

863

104

25062

40 - 44 

4008

5881

31

84

107

11188

2127

821

62

24309

45 - 49 

3855

4813

55

86

82

9929

2177

854

132

21983

50 - 54 

3590

4896

31

54

45

8632

2183

722

59

20212

55 - 59 

3363

4028

43

67

40

7366

2214

627

122

17870

60 - 64 

2912

3341

24

36

21

4792

1820

439

77

13462

65 - 69 

2625

2575

32

21

20

3182

1362

393

87

10297

70 - 74 

2091

1889

2

16

21

2244

1124

315

108

7810

75 - 79 

1530

1022

20

24

16

1316

785

253

81

5047

80 +

1107

519

12

1

69

700

611

152

118

3289

Total

52294

64824

360

1471

742

81922

39820

11831

1299

254563

Table 3. Sample size (number of journeys) of NTS 2003 by age group and transport mode.

Margin <10%

Margin < 20%

Margin ≥ 20%

 

 

 

We can thus see that for a margin smaller than 20%, a sample of approx. 1000 journeys is necessary, and for a margin smaller than 10% approx. 5000 journeys.

A margin smaller than 5% does not occur in any of the cells; this only occurs for the summing of age or transport mode. In some cases it helps to totalize a number of years.

 

In the Powerplay web application, the number of journeys can be made visible by, under 'Measure', selecting not the kilometres travelled, but the number of journeys (sample size, see also Table 3). The number of journeys used to calculate the mobility gives a rough indication of the expected margin.