The British National Travel Survey: Sampling and Weighting

1
The British National Travel Survey Sampling and
Weighting

• Kevin Pickering and Simon Holroyd
• National Centre for Social Research

2
National Travel Surveys

• Continuous survey in Great Britain (England,
  Scotland and Wales) from 1989
• Survey components
• Household interview
• Individual interview
• Vehicle section
• Travel diary
• Long distance journeys

3
National Travel Survey 2002

• Issued 15,230 household in Great Britain
  (England, Scotland and Wales)
• Two-phase design 684 PSUs, 22 addresses
• Issued sample spread over survey year - each
  quarter is nationally representative
• 7,437 fully responding (54.5)
• 1,576 partially responding (11.6)

4
Advantages of a continuous survey I

• Avoids headaches of a big bang survey
• Reduces learning curve for sponsor, survey
  organisations, interviewers
• Operational advantages (cost, time efficiency)
• Control for season effects (e.g. long distance
  trips for holidays, bicycle use, walks etc.)

5
Advantages of a continuous survey II

• Can aggregate years to allow analyses for
  sub-groups (e.g. regions)
• Easier to obtain funding when up and running
  more difficult for one-off surveys
• More attractive to survey organisations

6
Response rates

• Falling response rates - from 79 in 1991 to 65
  in 2001 (fully responding)
• Increased risk of bias in survey estimates
  associated with declining response rates
• Recommendation to develop strategies for
  improving response rates make use of partially
  responding households and to investigate
  weighting to reduce bias

7
Methods to improve response rates

• 5 incentives paid to each household member if
  household is fully responding - increased
  response rate by about 7
• Book of stamps sent with advance letter
• Make more use of partially responding households
  analysis of interview sample - all households
  with completed interviews for all householders

8
Weighting Interview sample

• Selection of household by interviewer at
  multi-occupied addresses (W1)
• Household non-response (W2)
• Exclusion of co-operating households at which not
  every householder completed the individual
  interview (W3)
• Generate composite weight (W1 x W2 x W3)
• Calibrate to age/sex and region pop ests

9
Weighting Fully responding sample

• W1, W2, W3 as for interview sample
• Exclusion of households that did not fully
  respond, but completed all individual interviews
  (W4)
• Generate composite weight (W1 x ... x W4)
• Calibrate to age/sex and region population
  estimates and characteristics from the (weighted)
  interview sample