Les témoins

Previously in this study, Table 2 listed the number of witnesses by sex in various age groups. Overall, 70% of the witnesses were male, 30% female. We shall attempt to assess the implications of this result and others from Table 2.

There is no source which lists exact data for the average number and sex of drivers and passengers in motor vehicles. It appears such information is not recorded. From statistics compiled by the National Safety Council 1979 edition of Accident Facts (1), one can calculate the relative number of miles driven by male and female drivers. Data covers the period from 1962-1978, a span containing over half of the EM events. In 1962, men drove 84% of the total mileage; by 1978, this figure had declined to 67%. The decline was steady, leading to the conclusion that an even higher percentage of total miles were driven by males prior to 1962. It should be noted that these percentages were derived on the assumption that men and women have equal chances to become involved in an accident. Additionally, the statistics refer only to drivers, not passengers. The average percentages for the 1962-1978 period are: males, 78% of miles driven; females, 22%. It was impossible to assign any possible error term to these figures, which are non-weighted averages.

Since the National Safety Council records only apply to the United States, the data from Table 2 for the United States and Canada (which is quite similar), for the 1960-1979 period, were extracted. Only drivers of motor vehicles were included. The results are in Table 13.

The overall ratio of male to female drivers -- 79% to 21% -- should be compared to that derived above from highway statistics, 78% to 22%. Given the uncertainties inherent in the calculations, the agreement is excellent. No selection factor for sex would appear to be operating, which is, prior to analysis, what might have been expected.

Using Accident Facts, we may calculate the expected percentages of drivers in each age group under the assumption that drivers of different age groups have equal chances of having an accident. This is clearly not true, as is known by all insurance companies and police departments. Drivers under about 26 are significantly more likely to be involved in an accident. This means that the calculated percentages for the Teenage and Young Adult groups will be inflated by some unknown factor. Table 14 presents the final figures.

Tableau 15 - Pourcentages attendus de conducteurs par groupe d'âge en Australie

Age	%
Adolescents	3,8 %
Jeunes adultes	13,5 %
Adultes	82,6 %

There is obviously a significant difference between Tables 13 and 14, but whether this can be attributed to selection factors, differences in driving habits, or the many assumptions inherent in this calculation, is unclear. In contrast to a previous paper (2), I do not believe a statistical comparison of these two tables is in order. For example, data from Australia yields the following results (3), which exhibit an even greater divergence from the EM data.

It would be best to simply conclude that teenagers might be overrepresented in the population of witnesses, and adults underrepresented. The causes may be (but are not limited to):

1. Larger relative numbers of teenagers driving at night and fewer adults
2. Larger relative numbers of teenagers driving in rural and isolated areas
3. A reluctance among adults to report their experience compared to teenagers
4. Selection factors of the EM event itself

Other studies have reported that the number of younger witnesses is higher than expected (4), so this result is not new. However, chances are that driving and reporting habits of witnesses can explain it without recourse to exotic explanations. To strengthen this assertion, consider the fact that, for U.S. cases, about 50% involved a single witness. The U.S. Department of Transportation reports that one-occupant trips, on the average, represent 50.2% of all trips. Average occupancy is 1.9 passengers per vehicle (5). These figures so closely match the EM data that it is prudent to conclude that no selection is occurring on the number of witnesses per event, and that vehicles are affected at random by the phenomenon.

To further investigate the question of witness involvement, contingency tables were constructed comparing the location of an event to witness characteristics. Table 16 presents the data for witness sex.

Chi-square for this table is 4.4, which is less than the value for the .05 level of significance, 7.81. The null hypothesis--that the parameters of witness sex and location are independent--is supported.

For those witnesses for which both age and sex were available, Table 17 was constructed. The two similar categories or rural and deserted locations were collapsed for ease of analysis.

Calculations for this table must be done using Saunder's "Remarkability" statistic, as numbers in some cells would be too small to use other, more standard techniques. Refer to the Appendix for a full discussion of the definition and use of this statistic. For our purposes, any remarkability (or R) value over approximately six will be considered as evidence that the null hypothesis is not satisfied. R for Table 17 is essentially zero, definitely not a remarkable result. This means that the hypothesis that younger drivers tend to drive in rural areas at a higher frequency, hence possibly explaining their over representation as witnesses, is not supported. The characteristics of witness location and witness age act independently.

Not enough data is available to make a comparison of age versus time of day, so the possibility still exists that younger drivers are on the road relatively more often at night than adults.

A breakdown of the number of witnesses by sex for each hour of the day showed no deviation from chance upon inspection.

Since the sorting of cases by time of day has yielded new results, additional correlations were attempted. Table 18 presents the average number of witnesses per case by three-hour intervals of the day.

The average number of witnesses per case is 1.91. The standard deviations for the averages are such that it is evident that the distribution can be considered random. For example, for 12: 00 A. M. to 3: 00 A. M., sigma = +0.96; for 6: 00 P. M. to 9: 00 P.M., sigma = +1.42. However, the average for 6:00 P.M. to 9:00 p.m. is the highest, and corresponds exactly to that time when many families are out in their cars. The two rush-hour periods have the lowest averages, not too surprising given the prevalence of driving to work alone (6). Thus we can make some sense of these numbers.

Ces moyennes, taken together with the apparent non-selection of the number of single witnesses per event discussed above, strongly indicate that no selection factors operate with respect to the number of people in a vehicle during an EM event.

We have seen that EM events occur more frequently in rural locations. Since they also occur preferentially at night (defined here as 6:00 P.M. to 6:00 A.M.), a correlation was attempted between location and the time of an event to determine if rural events happened more often at night. Table 19 is that contingency table.

Due to low expected frequencies, Saunder's statistic will be used to analyze the data. R is again essentially zero (actually, slightly negative), not a remarkable result. This means that the hypothesis that EM events occur preferentially at night in some types of locations has not been supported. Time of the event and witness location act independently.

This result is perhaps not startling, but is new, in that previous studies of either general UFO cases or particular subclasses have not investigated the relation between these variables.

A final comparison was then attempted between the number of witnesses per event versus location. Table 20 gives the result.

Using remarkability statistics, we calculate that R = 5.8, which does not quite meet our criterion for rejection of the null hypothesis. This is, though, an indication that there is some non-randomness in the data. In particular, there is a higher proportion of single witness cases in rural and suburban locations than in deserted locations, or especially, urban locations. But we do not have sufficient evidence to reject the null hypothesis. Any variations that exist could probably be explained by sociological factors.

To summarize our analysis of witness data in various categories, no dependence has been found between either time of day or location and the number of witnesses, the sex of the witness, or the age. No factors appear to be at work "selecting" particular witnesses, except for the fact that many cases occur in rural, isolated locations. Only the relative proportions of witnesses by age might differ from that expected, but the result is somewhat problematical.

These facts are highly significant because rarely before has such detailed analysis of witness characteristics been compared to the location or time of an event, for a well-defined, high-strangeness subset of UFO events. Neither have even approximate calculations been available before to determine the "expected" frequency of witness involvement by sex and age, except for gross estimates. The fact then that very few, if any, correlations were found between these categories must be accounted for in any theory that would purport to explain EM events. The data can best be understood if viewed in the framework suggested by Westrum (7): "High-threshold (UFO) experiences are not related to sociological factors--except those factors which determine one's opportunity to observe." All indications are that witnesses to EM events are a random sample of the available drivers (vehicles) on the road, by time of day and location.