Editor’s Note –

This article, ‘A Principal Components Analysis of 59 Variables Descriptive of Uncovered Spies,’ was originally presented as a paper that was read by its author at the annual meeting of the International Military Testing Association meeting that was held in San Diego, California on 28 October 1992. It did later appear in that organization’s annual convention monograph. However, due to the extremely limited distribution of the particular monograph, which incidentally was and is (and has been) also very difficult to obtain, the article’s author surprisingly did receive, over the years, quite a few requests, from all over the world, for ‘reprint’ copies. For some unknown reason, in the early months of 2002, a sudden increase in the number for requests for reprints was noted. Because of the still apparent interest in the article’s subject-focus, it was decided to ‘republish’ the article in electronic journal form.

It should be noted that the article was originally presented in 1992 (just about ten years ago) and, as a result, the described analyses were associated with a database that, since that time, have grown significantly larger. Back in the 1991-1992 timeframe the traitorous spy database was descriptive of 186 spies; actually at the time when the paper was read at the annual meeting the database had grown to include 197 spies. At the present time it now includes information pertaining to over 250 spies. Also, at the present time, more quantified descriptive variables are now present that were involved a decade before. However, with this said, it is believed that this article still is relevant in the overall study in the field of espionology. It should be noted that the ‘References’ section has been updated to the present time.
 
 

     In 1987, this investigator started a database for all those U.S. citizens who, since the end of World War II, have been uncovered spying on their own country. From the very beginning, it was decided that only quantifiable descriptive variables would be employed in constructing this database. The primary reason for wanting to build such a database was due to the fact this investigator discovered that no such database existed in the private sector and it also appeared then that such a quantified database (which could be subject to statistical analyses) did not seem to exist, even within governmental agencies and bureaus. To anyone not overly familiar with research conducted on the general topic of uncovered spies, it comes as a shock to learn that no one has apparently produced any systematic scientific research on this subject that has been published or even distributed on a limited basis. Generally, there have been a number of publications which accomplished nothing more than providing narrative-type descriptions of a limited segment of the known uncovered spies; the best previous attempts at quantification have occurred only in a couple of publications (e.g., Crawford, 1988; Jepson, 1987). In these, the only resulting numerical analyses reported were entirely limited to a very few frequency-count tables of matters such as amount of education, number of years of federal or defense-associated employment, age at time when caught, etc.). Not even simple descriptive statistics were used in these publications. In contrast, the present investigator has completed more than several investigations that have employed complex statistical analyses, some of which utilized so-called multivariate mathematical-statistical procedures (e.g., Stone, 1991a, 1991b, 1993a, and 1993b). 

     When the presently reported research was begun, the U.S. spy database involved 186 spies, each with quantified listings on up to 68 different variables. The choice of variables was mainly dictated by noting what kind of information was available in books, magazine articles, newspaper accounts, Government published security training monographs and the like, information obtained from Federal agencies/bureaus based upon Freedom of Information Act requests, and information derived from commercially available journalistic databases. Some of the variables in this database were psychological measurement estimations (e.g., intelligence quotients) based on estimation procedures which seem to be well-established in certain applied areas of psychology (Wilson, Rosenbaum, Brown, Rourke & Whitman, 1978; Stone, 1993a). The database is one that is certainly not static; it is frequently being added to, either with new variables or more often with the missing data-points being replaced with recently obtained information. It is most unlikely that this particular database will ever be considered as being final and complete. A number of the investigations, which have involved utilization of this database, have been reported in a review article (Stone, 1991a) that initially received a good deal of attention. 

     Since the original decision, as to whether to include a variable in this database, has almost always been more determined by the matter of information availability rather than anything else, it becomes particularly important to understand what basic aspects of spies are really being represented by the variables in this database. High redundancy was known to be a problem with a few of the involved variables. It was wondered whether factor analytic or principal components methodology might serve as a means to better understand the particular composition of variables included in the database. Would the most major components (or factors) be interpretable? Would there be some factors/components that were primarily represented by only a single variable? Could factor/component scores be used to replace some of the redundant variables’ measurements? Basically, since this particular database was the first known, fully-quantified one constructed to be descriptive of caught or uncovered spies, a factor or principal components analysis of same was the focus of a good deal of curiosity as no such previous accomplished analyses of this kind of content have been reported elsewhere. 
 
 

     The particular correlation matrix that was to be submitted to a principal components analysis was originally 68 x 68; however, it was decided, due to a rather major incomplete data measurements problem, to actually analyze 59 of the involved variables. The nine variables, not included in the analysis, were omitted as the numbers of measurement observations on each were less than the total number (i.e., 68) of variables in the database. For most of the analyzed 59 variables, the number of measurements involved with each variable generally differed somewhat, but most were fairly close to the total number of spies described in the database (i.e., 186). 

     Correlation coefficients in the analyzed correlation matrix ranged from -0.99 to 0.98. The employed computerized principal components routine made use of a variation of the rather old Jacobi algorithm to obtain trial eigenvalues. The number of principal components extracted and retained was set to be limited to the number of eigenvalues that had values in excess of unity. The resulting number of principal components thusly obtained was 17. It is interesting to note that use of Cattell’s scree test produced a graph that suggested consideration of only a five-factor/components limitation. However, for subsequent rotation of the components, this particular limited number of factors of components was not used. The 17 principal components were rotated using Kaiser’s varimax normalized procedure (the rotated loadings for these principal components are given in Table 1; the variables have been sorted into order according to principal component loadings sizes). When the number of variables is relatively large (i.e., larger than 40 or so), results obtained using a principal components solution usually can be expected to closely resemble those obtained based on most of the other well-known orthogonal factor analysis solutions. Note - Table 1 is found at the end of this article, following the References section.

     Inspection of Table 1 reveals that the 17 rotated principal components totally accounted for about 81% of the common variance for the involved 59 variables. If one limited the number of principal components to only the first five, the number suggested by employment of scree graphing, these five are seen to account for only about 40.4% of the common variance, or about one-half of that accounted for by the full 17 principal components. The first principal component, the largest one, accounted for about 13% of the total common variance and about 16% of the common variance accounted for by the full number of 17 principal components. The communality values range from a low of 0.586 up to 0.990. The mean communality value was 0.770. 
 

     Since the present investigator is familiar with what the variables represent, it was not difficult to suggest a label for the first principal component, which clearly was bipolar in form, and this was "Monetary vs. Ideology Motivations for Spying." In previous reports, some of the database variables, such as Foreign Preferences, Jewish Background, Year in Which Spying Started, Birth-year, Having of Money Problems, etc., have already been noted (e.g., Stone, 1991b, 1993b) to be highly correlated with a similar derived motivational variable. The second principal component can be rather easily labeled as a sort of "Mental Ability" construct. The third component looks to be an "Age/Experience" matter. The fourth component looks like a bipolar "Military vs. Civilian Background" construct. A label of "Native-born Origins" seems appropriate for the fifth component. "Substance Abuse" represents a very clear title for the sixth component. The seventh component seems to justify a name of "Length of Sentence Given for Spying." "Sexual Misconduct" is a likely title for the eighth component The ninth component is not really easy to name; however, it, when some interpretation is forced, looks a bit like "Air Force Spies (but not Navy Spies) Described in Government-Produced Publications." The tenth component is also not easily interpreted; one can impose a label of something such as: "Army Spies Arrested Overseas by a Foreign Power Agency." The eleventh component seems easy to understand; it rather clearly can be defined as a "Mental and Emotional Problems Involving Criminal or Deceptive Actions" kind of construct. The twelfth component can be interpreted as a sort of "Coerced into Spying (and Not Having Financial Problems) to Benefit Countries Other than the Soviet Union" matter. "Long-term Spying Ending in Suicide" seems to be an appropriate name for the thirteenth component. Component fourteen can be understood as a sort of "Disaffection Motivation as the Basis for Spying (and Being in the Army)" matter. Component fifteen can be labeled as a kind of "Criminally Based Volunteering to Spy (More Frequently Found in the Air Force)" type factor. "Intelligence Community Background Associated with Previous Security Violations" can be used to describe the sixteenth component. The seventeenth and last extracted component can be labeled as a "Black-American Marines in Overseas Locations" matter. 
 
 

     It is encouraging to note that most of these extracted and rotated orthogonal principal components can be somewhat readily named and seem to be perhaps even interpretable. Also noted is that each of these components has more than one of the variables leading high on them; none of the components seems to be just a single-variable kind of factor. Interpretability seems not to decline much even after the fifth component (which, according to scree charting, seemed to suggest that it be considered as the final one). 

     Inspection of the ‘larger’ extracted and rotated principal components seem to support a type of belief that the studied spy database does seem to rather include measurement of at least three or four different varieties of motivations for spying. The first and most major component or factor was rather easily identified as being representative of a bipolar motivational construct, "Monetary vs. Ideology Motivations for Spying." This particular construct also very clearly emerged in another analysis of the database, when it was seen following cannonical correlational analyses of portions of this spy database (Stone, 1991b, 1992, and 1993b). The present analysis seems to suggest that the other two spying motivational variables (Coercion [i.e., Component 12] and Disaffection motivation [i.e., Component 14]) can be regarded as being orthogonal rather than as bipolar representations on a single continuum, that has been suggested in a two-factor motivational theory for spying (Stone, 1993b). Actually, when developing this noted particular two-factor theory, I was quite aware that the Other (which includes ‘coercion’) and the Disaffection motivations were somewhat separate and that forcing them into being bipolar ends of a single dimension was mainly for the purpose of being able to describe a less complex motivational theory. It should be understood that the actual number of spies whose primary motivation for spying was Other or Disaffection was quite small; a very great majority of the studied spies were known to have spied because of greed or ideology. 

     The ninth principal component, although somewhat difficult to identify or interpret, does have some interesting connotations. It seemingly represents a component involving spies from one particular military service (and excluding another particular military service) whose cases tend to be described in publications produced by the Federal Government. It is interesting to contemplate what possible governmental policies such a component might represent. A single underlying basis for this component might simply be the government-published book, authored by Crawford (1988). Familiarity with the building of this database does cause its developer to believe that the U.S. Navy historically has been prone to only provide a very minimum of information regarding its members who have been uncovered as spies. In fact, it appears that it has been Navy policy to provide no more information to the public, if any at all, than it might be required to do. 

     Based upon these principal components, component scores have been computed for each of the 186 spies in the database. These 17 new sets of scores have been used to increase the number of variables in the database. This of course resulted in an increased redundancy of measurement within the database. 

     The principal components analysis, that has been accomplished with the caught U.S. spy database, has provided some new insights and understandings regarding the database itself. In the future, when new measurement variables are to be considered for adding to those already employed, the potential new variables can be tested to discover whether they might be adding new kinds of information or whether they might be adding additional measures of some of those constructs for which measurement already exists in the database. However, what is viewed as being the single most important matter in the accomplishment of the principal components analysis of the database is simply the fact that the database possessed sufficient measurement quantification to warrant such an exploratory, but sophisticated, statistical analysis. Of any known existing caught-spy databases, this particular one seem to be the only one that allows for such systematic analyses. This kind of situation brings to mind the comment once made by Lord Kelvin: "Until you have measured it, you don’t know what you are talking about." 
 

Crawford, D. J. (1988). Volunteers: The betrayal of national defense secrets by Air Force traitors. Washington, DC: U.S. Government Printing Office. 

Jepson, L. P., Jr. (1987). Espionage directed towards U.S. military interests since World War II. Washington, DC: Counterintelligence Division, Directorate for Security and Counterintelligence, Defense Intelligence Agency. 

Stone, L. A. (1990). A study of spies who were caught but never punished. Bulletin of the American Academy of Forensic Psychology, 11, 10. 

Stone, L. A. (1991a). I spy a myth. Security Management, 35, (10), 26-32. 

Stone, L. A. (1991b). The use of demographic and biographic information to predict specific spying motivations. Unpublished manuscript. 

Stone, L. A. (1992). Canonical-correlation between security-clearance adjudication concerns and later motivational causes for espionage behavior. Forensic Reports, 5, 

305-316. Also as an abstract in the (1991) Bulletin of the American Academy of Forensic Psychology, 12, 10. 

Stone, L. A. (1993a) Intelligence levels for U.S. citizens identified as spies. Read at the American Psychological Association meeting, Toronto, Canada, August 1993. 

Stone, L. A. (1993b). A two-factor motivational theory for spying behavior. Read at the Psychonomic Society meeting, Washington, DC, November, 1993. 

Wilson, R. S., Rosenbaum, G., Brown, G., Rourke, D., & Whitman, D. (1978). An index of premorbid intelligence. Journal of Consulting and Clinical Psychology, 46, 1554-1555. 
 

Note - in the following couple pages, Table 1 is depicted. Due to the unusually large size of this table, it was necessary to divide the table into three sections. For anyone wanting to carefully examine/study the numerical entries in this large table, it is suggested that the three electronic journal pages, in which this table are shown, are printed to a hard copy format. With the three print-outs, they can be ‘cut’ and arranged so a to create a single hard copy table that resembles the original one with its proper ‘oversized’ dimensions made less confusing.

Table 1 normally should be shown here, however it was found that it simply would not 'fit' as it is so large.  It was decided that readers who are indeed highly interested in the results of the presently described research, that they could email or 'snailmail' the author and request a hardc-copy of Table 1.  Sorry for this problem.

Dr. Stone's mailing address is:  Dr. LeRoy A. Stone
                                                     P.O. Box 395
                                                     Harpers Ferry, WV   25425

His E-mail address is:  lastone2@earthlink.net