Science -> Homeclick!

Cette page en franšaisCliquez!

Science and the UFO Phenomenon:

Dr. S. R. Saunders, Extrinsic Factors in UFO-Reporting:


Dr. David R. Saunders, Professor of Psychology at the University of Colorado and later Dr. David R. Saunders is a Research Associate at the University of Chicago's Industrial Relations Center.

More on Dr. Saunders here.

Dr. Saunders.

More UFO science papers here.

AIAA 13th Aerospace Sciences Meeting
Pasadena, Calif. / January 20-22, 1975
AIAA paper 75 - 43

Extrinsic Factors in UFO-Reporting

D.R. Saunders
The University of Chicago
Chicago, Illinois


Using the method of stepwise multiple correlation analysis, five factors are identified having empirically-demonstrable effects on the production of UFO-reports; other factors have been simultaneously set aside as irrelevant. In order to maximize the number of reports, it is helpful (a) to assemble a large number of potential witnesses, (b) to educate them at least through high school, (c) to station them where they can see (d) to give them a place to report, and (e) to provide one or more examples of such reports. Data on these factors alone suffice to provide a multiple correlation of 0.82 with actual numbers of UFO-reports produced in US counties, and they come at least very close to accounting for the statistical reliability of this criterion. Several hypotheses predicated on alternative models of the UFO-reporting process are affirmatively rejected by the data reported here.


Irrespective of the eventual disposition of UFOs themselves, it is an unmistakable fact that UFO-reports do exist. By August 15, 1947, as one result of a major wave of UFO-reports peaking in July of that year, Gallup (4) was able to report that 90 percent of the American public had heard of "flying saucers." And by the spring of 1966, extrapolating from new survey data, Gallup (5) estimated that more than 5 million Americans would claim to have seen a ''flying saucer.'' Independent data gathered by Lee (6) in 1968 also suggest the existence of several million potential UFO-reports. Quite obviously, flying-saucer-seeing has been a widespread phenomenon.

However the bulk of these potential reports are inaccessible. Lee's data indicate that only about one witness in ten even tried to render a UFO-report in the first place, either to the local press, the local constabulary, or the local military. These recipients, in turn, have apparently rejected 80-90 percent of their UFO input, and have created a record that accounts for only about 1 percent of the estimated original population of potential reports. It would seem foolish to assume that the remnant 1 percent is a random sample from the larger population, but it is still a large body of material that may be analyzed on its own merits.

A variety of speculative hypotheses have been advanced for UFO-reporting behavior, and some of these have been given wide circulation without having been in any way tested against data. The primary purposes here are (1) to demonstrate one way in which such tests may be carried out and (2) to present the resulting substantive findings.

Method and Procedures

The basic method used in this study is stepwise multiple correlation. The criterion variable was the frequency of recorded UFO-reporting for each of a series of well-defined demographic units. This total frequency was also broken down according to a rudimentary classification of types of reports. The predictor variables were various other measurable attributes of the demographie units, such as population, area, etc. The demographic units themselves were the counties of the United States (*), which were regarded as a sample from a statistical population of possible counties.

(* The exact number of recognized counties varies slightly from census to census. The analyses to follow were based on 3053 counties which were consistently defined in 1950 and 1960 and for which complete data could be assembled. Approximately fifty counties, including all of Alaska, were excluded by these criteria.)

The main analysis whose results will be discussed in detail was approached in several stages. In the first stage, only three predictors were considered, and these were examined only in relation to a criterion measure based on a relatively limitad collection of UFO-reports. Since these results were encouraging, additional predictors were developed successively, until there were fourteen predictors altogether. At the same time, the collection of UFO-reports was steadily enlarged, leading to more reliable determinatian of the reporting activity of each county and, in the final analysis, to a fourfold breakdown of that activity.

It was anticipated, and verified in the first stage analysis, that the best results would be obtained by transforming the raw variables of the study prior to their intercorrelation. The standard multiple regression model postulates additive contributions from each of the independent predictor variables. However, if this additive model is applied to the logarithms of the raw variables, the effect is to fit a model in which each raw predictor actually makes a multiplicative contribution. Such a multiplicative model seemed conceptually appropriate for the data to be analyzed, and its use led to substantial enhancements in the zero-order correlations. For further discussion of this issue, see Bartlett (1) or any textbook on econometrics.

The fourteen predictor variables became available to this study in the following chronological order:

Considered separately, each of these six predictors is positively and significantly correlated with the criterion of UFO-reporting. However, when they were first tried in combination, X1, X4, and X5 were all found to be completely redundant with other variables. X2 and X6 emerged as the major predictors, both with positive weights. X3 emerged as a marginal predictor, still statistically significant but with less weight than it had after the first stage analysis. The next two predictors were now included with the hope of forcing X3 to nonsignificance, as had already been accomplished for X1.

X7 - Longitude of the county, measured in degrees west of Greenwich. Specifically, this was taken as the coordinate of the county seat as given by the Times of London World Atlas (10). This was correlated without any further transformation.

X8 - Latitude of the county, measured in degrees north of the equator. The comments under X7 apply.

Both of these new measures yielded positive zero-order correlations with the criterion. In multiple correlation, X8 was completely redundant, while X7 behaved as a 'parallel form' of X3 - while overlapping appreciably, neither X3 nor X7 was able to supersede the other. Two more simple-to-define geographical measures were added next.

X9 - Proximity to salt water. This variable was scored 1 if the county is adjacent to salt water (an ocean or the Great Salt Lake), and otherwise 0.

X10 - Proximity to fresh water. This variable was scored 1 if the county is adjacent to (or contains) a large natural body of fresh water, and otherwise 0.

Both of these new measures yielded positive zero-order correlations with the criterion, and each provided a small but statistically significant increment to the multiple correlation. These results supported the idea that an important geographical variable exists, but also indicated that this variable had not yet been found. The remaining four measures were added to the study all at one time, but for varying reasons.

X11 - Proportion of population over age 65, according to the 1960 census tabulation (3). For purposes of correlation, this was transformed to the logarithm of (1-p)/p, effectively reversing the variable. Thus, positive correlations may show an effect of (comparative) youth. No measure of age had been included among the previous predictors.

X12 - Proportion of population with at least five years of education, according to the 1960 census tabulation (3). This was transformed to the logarithm of p/(1-p). In view of the consistent importance of education through the previous stages of analysis, it seemed desirable to explore at least one more variant of the education measure. Since variables X6 and X12 are correlated only 0.816, they evidently do reflect different aspects of the educational process.

X13 - Number of distinct newspapers currently published in the county, as counted in a compilation prepared for this purpose from several sources. For purposes of correlation, this was taken as the logarithm of (N+1).

X14 - Number of newspaper editions currently published in the county per week, as counted in the same compilation used for X13. For purposes of correlation, this was also taken as the logarithm of (N+1). Since variables X13 and X14 are correlated only 0.856, they apparently do measure different aspects of the availability of newspaper coverage.

Once again, all the new measured yielded positive zero-order correlations with the criterion of total UFO-reporting. The complete matrix of zero-order correlations for all fourteen predictors is included in Table 1. This table also gives the correlations between these predictors and five variants of the criterion variable; these must now be defined.

Y0 - Total UFO-reporting. All the variants of the criterion variable are based on the UFOCAT catalog of UFO-reports, and were tallied from a version of this catalog containing 59237 total entries (10). This total count included numerous duplicate reports of the same events, as well as many reports of events occurring outside the usable counties of the United States. Thus, only 18122 entries were used to generate the criterion measures now being described. When these 18122 UFO-reports were distributed among the 3053 usable counties, the number of reports per county ranged from 0 (numerous instances) to 598 (Montgomery County, Ohio - the home of USAF Project Blue Book). For purposes of correlation, these counts were transformed to the logarithm of (N+1).

Each of the 18122 usable UFOCAT entries is also classifiable as to the 'Type of Event,' on a scale which roughly characterizes the 'strangeness' of the report. These types may be indicated as follows:

Type 0 - Not reported as a UFO.
Type 1 - Reports of non-moving objects.
Type 2 - Reports of continuously moving objects.
Type 3 - Reports involving motion with one discontinuity.
Type 4 - Reports involving motion with multiple discontinuities.
Type 5 - Reports of ob@ects entering the witness' frame-of-reference
Type 6 - Landing reports.
Type 7 - Occupant reports.
Type 8 - Contact reports.
Type 9 - Reports involving post-encounter effects.

More complete definitions of these types of report are given in the UFOCAT Codebook (10). For purposes of this study, Type 0 reports were completely excluded and the remaining types were grouped into four broader categories, as follows:

For purposes of correlation, each of these sub-criterion measures was transformed to the logarithm of (N+1).

Before taking any logarithms, it is true for each county that Y0 = Y1 + Y2 + Y3 + Y4. However, in scanning the UFOCAT file for the purpose of generating these criterion data, a restriction was enforced so that no more than one event per county per day could be counted toward Y0. If more than one otherwise countable entry was found in the file, only the one with the highest code for Type of Report was used. This procedure introduces a small negative bias affecting the correlations between the Y's. This procedure avoids a larger problem associated with differences in reporting practices of different sources underlying UFOCAT, since some of these sources do and some do not tend to detail a cluster of reports from one county on one date.


The main results of this study are contained in Table 2, which lists the predictor variables in the order of their selection for each of the five possible multiple correlation runs. Each list stops with the last predictor having a remarkability (*) exceeding 5 bits. The listed values for 'partial r' are obtained by holding all previously chosen predictors constant, but the 'multiple r' includes the contribution of the predictor listed on the same line.

(* Remarkability is net information favoring rejection of the null hypothesis. The calculation of remarkability allows for the explicit selection effects involved in basing the stepwise regression on the best available predictor at each step, as well as for implicit selection of the sign of each partial correlation and for sample size (11). On each line of Table 2, the odds are 2ER to 1 against obtaining such a large partial correlation from chance alone.)


1. Population, vintage 1960, is the only predictor variable to be chosen in all five of the stepwise regression solutions, and in every case it is the first predictor to be chosen. Almost every UFO-report involves at least one witness. It comes as no surprise, therefore, that the counties containing more potential witnesses have produced more reports. The magnitude of this correlation is undoubtedly enhanced by the wide range of populations represented in our sample of counties, the ratio of the largest to the smallest being well over 10,000 to 1. The 1960 population data work better than the 1950 population data because 1960 is more nearly the midpoint of the period during which the bulk of the UFO-reports were produced - 1947 through 1972. When 1960 population is held constant, the partial correlations between 1950 population and the criteria are consistently negative; this may be interpreted as a positive relationship between growth-rate and UFO-reporting, but one that disappears as soon as educational effects are accounted for.

2. Education, measured in terms of the proportion of adults who have completed high school, is the second most powerful predictor found in this study; it is chosen immediately after population in four of the five analyses, but it is not chosen at all to help predict Y4. Whenever it is chosen its weight is positive, that is, more reports per capita are produced by the better-educated counties. The differential role of education in predicting the subcriteria is quite interesting. Education appears to be maximally relevant in connection with Y2, which is the subcriterion for the strangest remote reports. On the other hand, education is simply immaterial in relation to Y4, which is the subcriterion for the strangest nearby reports. All these effects could be anticipated, however, starting from the simple-minded assumption that UFOs have physical reality.

Education, as measured by completion of the 5th grade, is unable to contribute anything to these analyses that is not already better contributed by the 12th grade education measure. Evidently, if there is any threshold effect associated with the contribution of education to UFO-reporting, the threshold is not below the 12-year point. The census volumes do not contain any data with which to explore other possible thresholds.

3. Area is the third most powerful predictor found in this study, appearing in four of the five analyses: subcriterion Y4 is again the exception. The contribution of area is always positive, that is, at any particular level of education, more reports per capita are produced by bigger counties. If it is assumed that UFOs have physical reality, this could simply mean that there are more of them waiting to be seen in bigger counties. A second possible explanation for this predictor is that area is merely acting as an indirect measure of seeing conditions; large counties in the US do tend to be associated with desert areas, where visibility is relatively good. Poher (8) has displayed a correlation between seeing conditions in France, measured as the number of hours of sunshine per year, and numbers of UFO-reports per department. Both longitude, which was a meaningful predictor in some of our preliminary analyses, and proximity to salt water, which now appears in all the same analyses as area, may also be interpreted as indirect measures of seeing conditions.

4. If we regard X13 and X14 both as measures of the availability of newspaper coverage for a UFO-report, we may observe that one or the other of them does appear in four of the five analyses. Again, the weights are always positive, and the general interpretation is easy. The vast majority of UFO-reports that have been collected by UFOCAT were originally published in newspapers, even though UFOCAT's immediate source for a report is typically some secondary or tertiary publication. Thus, we might easily argue that the contributions of X13 and X14 are simply artifacts of the process used to collect the criterion data. This would be true, but the argument really applies equally to all the measures used in this study; this is why we refer to them collectively as 'extrinsic factors in UFO-reporting.'

The differential validity of X13 and X14 is interesting, and provides a further indication that the breakdown of the criterion is meaningful. The pattern of relations suggests that reports classified here as Types 1-4 are being treated by the press as 'filler', which may or may not be published depending on the availability of space. The more editions that are being published in a county, the more likely such space is to be available. On the other hand, reports classifled here as Types 5-9 are 'news' in the full journalistic sense; space can always be found for these reports, but they will be first be subjected to a much more careful editorial review, so that their likelihood of actual publication is primarily a function of the number of editors who have an independent opportunity to say Yes. The latter effect is particularly pertinent for Types 7-9, which underlie subcriterion Y4.

5. Only two other predictors appear anywhere in Table 2. These are income and race and they appear only with marginal significance and only in connection with Y0, Y1, and Y2 - the subcriteria based on the largest amounts of data. It seems most likely that these measures are acting as indirect substitutes for one or more other variables not included in the study. Perhaps the most important thing to note is that any direct contributlon that these variables may make must still be positive, that is, it is the 'good' end of each of these variables that contributes to increased UFO-reporting. Nowhere in this study do we find even an indirect derogatory implication concerning the UFO-reporter. Nowhere in this study do we find any suppressor variable with sufficient remarkability to require our notice.

6. The largest multiple correlation reported in Table 2 is 0.806 for the prediction of Y0. The largest multiple correlation obtainable from Table 1 is 0.819, which may be obtained for the prediction of Y2 when Y1 is included as the first predictor. The latter compares with only 0.774 which is available for the prediction of Y2 from the extrinsic predictors alone. Similarly, Y3 and Y4 are better predicted by Y2 alone than they are by any combination of the extrinsic predictors - for Y3, 0.589 (Table 1) is greater than 0.548 (Table 2), and for Y4, 0.360 (Table 1) is greater than 0.330 (Table 2). Each of these differences is an indication that there is more reliability in the criterion than we have accounted for with the predictors. The magnitudes of these differences suggest that one or even two important sources of variance remain to be discovered; alternatively, there may be a much larger number of individually less important missing predictors.

7. On the basis of an examination of the French landing wave of 1954, Vallee (14) has proposed that what he calls Type I reports are most common in regions of low population density, and that this is a sign of intelligence on the part of the UFOs. Type I on Vallee's scale is almost identical with Types 5-9 on our scale of strangeness; actually, the French cases considered by Vallee as supporting his hypothesis are mainly Types 6 and 7. If population density is population per unit area, then Vallee's hypothesis predicts a negative weight for population and/or a positive weight for area under the conditions of the present study. The population aspect of this prediction is strongly rejected by our results, but the area aspect cannot be unambiguously rejected. However, if area in the US is really acting as an indirect measure of seeing conditions, then it will disappear from the stepwise regression analysis as soon as an adequate measure of seeing conditions is provided; in such an event the area aspect of Vallee's hypothesis will also be rejected.

In view of the relatively small variance associated with either the populations or the areas of the French departments (leaving Paris and its suburbs out of consideration), as compared with the populations and areas of the US counties, it does not seem likely in any case that Vallee's hypothesis can hope to provide a sufficient explanation for the observed distribution of French reports in 1954. Even if area is retained as a predictor in its own right, its weight is not big enough to do the job. Noting that X13 rivals X2 in its zero-order validity for Y4, and that it makes a demonstrable independent contribution to the prediction of both Y3 and Y4, it is reasonable to suspect that X11 may be the crucial agent leading to the data interpreted by Vallee as an effect of population density. This explanation would simply require that the number of newspaper editors per capita or per square kilometer be substantially higher in rural France than in metropolitan France.

It may be observed, of course, that the inability to support Vallee's hypothesis of low population density is a function of the use of a relatively coarse geographic grid (whole counties). Since we are aware of no report that a UFO has landed on the same precise spot where a person was standing, it could be argued in the limit that all landings are in places with zero population density. The interesting problem, then, is to determine on how large a scale Vallee's principle can still be supported.

8. On the basis of a reinterpretation of Gallup Poll data, Warren (15) has proposed that 'status inconsistency,' a concept that he attributes to Lenski (7), is directly operative in the production of UFO-reports. Perhaps the strangest thing about Warren's paper on this subject is that it is totally devoid of any attempt to assess statistical significance, even though it is identified with a laboratory and published in a journal which are ordinarily quite statistically sophisticated. When this deficiency is corrected, it becomes apparent that Warren's data provide no more support for his thesis than could reasonably be expected from tables of random numbers. Certainly, there is nothing in Warren's published data which necessarily motivates such a complicated theory of UFO-reporting as the theory of status inconsistency. The non-scientific ramifications of this situation need not be explored here.

Nevertheless, it is at least an interesting methodological exercise to look for effects in the present data that might be attributed to status inconsistency. Reduced to its essentials, Warren's theory argues that interaction effects corresponding to combinations of the present variables X4, X5, and X6, are primarily responsible for UFO-reporting. We have already seen that X6 (education) does play a major predictive role in its own right, whereas X4 (race) and X5 (income) make little or no independent contribution. We may look for the interaction effects simply by trying to use the product-terms X4X5, X4X6, X5X6 and X4X5X6 as additional predictors in the multiple correlation; status inconsistency will be supported if a significant negative weight appears for any of these product-term predictors. In effect, status inconsistency proposes that X4, X5, and X6 act as moderator variables (9) for one another, with particular sign relationships.

When the indicated product-terms were computed and tried as predictors, no significant multiple correlation improvement associated with a negative weight was produced for any of the five criterion variants. However, the product-term X5X6 did yield a positive weight in all five analyses and was supported by as much as 17.0 bits of remarkability (in the prediction of Y1); thus, status consistency with respect to income and education may be a meaningful positive predictor.

Again, it may be argued that the theory of status inconsistency is supposed to apply to individuals and not necessarily to averages for whole counties. For this reason, this test of the theory was not a very crucial one. Nevertheless, it remains true that there is no empirical support for this theory at either the county or the individual level.

9. Also basing his reasoning on the French UFO wave of 1954, Toulet (13) has proposed an epidemiological model to account for the numbers of reports yielded by the various departments. The important feature of this model is that reporting is assumed to be facilitated by the existence of other reports; the exact mechanism of facilitation proposed by Toulet is a secondary feature. While his analysis does employ certain simplifying assumptions, Toulet does adduce supporting data which are incompatible with the simpler hypothesis that reports are independently produced.

It is possible to analyze the data of this study in a way which displays a similar autocatalytic effect. This was accomplished by supplying the power-terms, X2E2 [= X2X2] and X2E3 [= X2X2X2], as possible predictors along with the other fourteen extrinsic measures. Given the choice between X2, X2E2, and X2E3, the stepwise algorithm invariably chooses X2E3 as the best single predictor of UFO-reporting; then X2 and X2E2 either do not appear, or appear later with negative (suppressor) weights. X2E3 is enough better than X2 to account for about one-half of the previously unpredictable reliability of the criterion, which was discussed in paragraph 6. Also, when X2E3 is used in place of X2, even the marginal utility X4 and X5, which was discussed in paragraph 5, disappears. The net effect is an appreciably higher coefficient of multiple correlation based on a smaller number of extrinsic predictors. However, the multiple correlations that were available by including the other subcriteria as predictors are not enhanced by X2E3; the ceiling is still 0.82.

The effects just described are strongly remarkable, and provide compelling evidence for a curvilinear dependence of UFO-reporting on population; the number of reports is a positively accelerated function of the population, which is consistent with the facilitation of some reports by other reports. There is still more than one way to conceptualize this pattern. Toulet, who borrowed his mathematical model from the field of public health, writes as if he were discussing the contagious process of a mental illness. Perhaps so. Alternatively, perhaps we are merely observing the reluctance of a UFO-witness to report his experience until another witness has 'broken the ice.' Or perhaps we are observing an effect of non- random sampling in which the same witness is more likely to make a report if he has already made a previous report. These possibilities cannot be distinguished by the present data.

10. One obviously important factor that this study has neither controlled nor assessed is the effect of local UFO- investigative groups. The most extreme example of this effect is provided by Montgomery County, Ohio, where the well-publicized local presence of Project Blue Book has elicited several times the number of UFO-reports that would otherwise be predicted for such a county. The effect of any such local enterprise will be to enhance the correlations between the subcriteria without enhancing their extrinsic predictability. It seems entirely possible that this effect is of sufficient magnitude to account for the remaining unpredictable reliability of the criteria used in this study.


Valid XHTML 1.0 Strict

 Feedback  |  Top  |  Back  |  Forward  |  Map  |  List |  Home
This page was last updated on May 26, 2005.