The impact of seemingly minor methodological changes on estimates of travel and correcting bias.

Abstract

In 1998, evidence was provided suggesting that an apparent 15% decline in domestic travel estimated from the Canadian Travel Surveys (CTS) of 1994, 1996, and 1997 was possibly misleading. This decline could actually be the result of methodological changes in the survey rather than a change in travel behaviour. Here, realistic assumptions and a simple computational method are utilized to demonstrate that change in bias may account for the drop in estimated trips. The analysis depends on a trip recall salience scale. Use of this scale allows correction for an under-reporting bias arising from reduced trip recall rates for low-salience trips. When the 1996 and 1997 CTS data are corrected, estimates of change in total trips are near zero. This finding corresponds to perceptions within the tourism industry. The bias correction methodology is applicable to other large surveys involving recall of past behaviour in which multiple events are recalled.

Cognitive Processes in Hunters' Recall of Participation and Harvest Estimates

“…Bias can also occur when respondents approximate answers by rounding estimates to numbers ending in zero or 5 (Hultsman et al 1989, Vaske et al 1996, Beaman et al 1997. In a situation where participation or harvest was actually 23, the response may be 20 or 25.…”

“…For example, if days of participation was 7, the multiplier use neighborhood was potentially ±3 around a multiple of 7. As 0-5 heaps have an unknown but potentially large influence on the distribution of responses in multiplier neighborhoods (Vaske et al 1996, Beaman et al 1997, Vaske et al 2003, 0-5 responses were not included in multiplier neighborhoods. In the days = 7 example, if there is no preference for a multiplier of 2, about 1/7 of the responses for the neighborhood would fall on 14 if it were not for 0-5 heaping on 15.…”

“…These processes potentially yield responses that differ from what actually happened. Approximate responses are suggested by response heaps (e.g., spikes in the data for numbers ending in zero or 5) that occur because some responses are given more often than would be expected by chance (Sen 1973;Tarrant and Manfredo 1993;Anderson et al 1996;Vaske et al 1996Vaske et al , 2003Beaman et al 1998).…”

Comment on "Digit Preference in Reported Harvest Among Illinois Waterfowl Hunters" by Craig A. Miller and William L. Anderson

“…Chu, Eisenhower, Hay, Morgenstein, Neter, and Waksburg (1992) suggest the bias is quite large. Using PSHC could possibly reduce bias in total harvest for Illinois from 50% to 35%, from 30% to 15%, from 7% to 5%, or (Beaman et al, 2000), the need for a 15% bias correction was identified with changes in survey methodology considered unimportant at the time of the survey. Benchmarks that could be estimated were certain types of international trips by Canadians (Auctor, 1998).…”

“…The abrupt change in methodology means the use of the old data may be minimal. This is because one generally does not know what part of change in estimates between old and new ones is a result of change in bias and what part is a result of change in behavior (Beaman, Beaman, O'Leary, & Smith, 2000;Kunert, 1998). Is there a conundrum?…”