Here is Testing Self as Control System

Testing the Self As A Control System:
              Theoretical and Methodological Issues

     Robertson, R. J., Goldstein, D. M., Mermel, Musgrave, M. (n.1)

       Interest in the nature of the self has recently rekindled.
Epstein (1973) proposed that the self is a theory the organism
formulates about its own nature to enable it to make decisions
for action. Rogers, Rogers, and Kuiper (1979) viewed the self as
a "cognitive prototype". Hull and Levy (1977) considered it to
have an organizing function. All of those conceptions point to
the self as a component or agency of the organism involved in
regulating or controlling its behavior.

     Many research reports seem to rest tacitly or explicitly on
that view. Rogers, Kuiper, and Kirker (1977), Bower and Gilligan
(1979), and Kendzierski (1980) offered evidence that data are
encoded in memory more firmly when they relate to the subject's
self. Other studies have proposed that people modify their
actions in ways that maintain the self-concept (see reviews and
discus-sions by Bem, 1967; Pryor, 1980; Wegner, 1980; Carver and
Scheier, 1981).

     A theoretical problem, however, arises in assuming that the
self controls action: the problem of assuming the self's tangi-
ble existence without a theory in which it would be more than a
meta-phor or hypothetical construct. Though Epstein (1973)
proposed that the self is a theory the organism holds about
itself, the question remains as to the agent that constructs the
theory. Bandura (1978) tackled this problem by referring to
"cognitive structures that provide reference mechanisms...and
subfunctions for [executing] behavior." However, he declared that
the self- system merely "refers to" these cognitive structures
and he maintained that it "is not a psychic agent that controls
behavior," still leaving open the question of what the agent
might be.

     We believe that Bandura's conclusion is unwarranted. A good
theoretical argument can be made for the self as an agent. (see
Robertson and Goldstein, 1987.) We report here part of our
search for appropriate methodology with which the conception of
self as agent can be investigated. We will describe a series of
experi-ments that make use of W. T. Powers' (1973a, 1973b, 1978)
theory of the organism as a hierarchy of interlocking control

     Experiments by Marken (1980, 1986), Plooij and Plooij
(1986), Robertson and Glines (1985), derived from Powers' theory,
have shown that various kinds of overt action (for example,
tracking, infant attachment behavior, complex learning) require
explanations that postulate more than one level of control sys
tem. Since, in the hierarchy, control systems of a given level
receive their reference signals as outputs of the level above, an
obvious question arising from these studies concerns the analysis
of how concrete behavior is influenced by higher-level control.

     For example, in Robertson' and Glines's (1985), study of
complex learning, they observed incidentally that some subjects
controlled their approach to the task by their notions of the
kind of person they were. Their strategies for gaining informa-
tion to solve the task were circumscribed by a (higher priority)
necessity to maintain consistency of the self-concept. Similar-
ly, Carver and Scheier (1981) have argued that the findings of a
broad array of social psychology studies can be interpreted in
the same way.

     The present investigations were undertaken not simply to add
to the list of such studies but in the hope of developing a more
precise method that can deliver better results than mere statis-
tical significance. If a person has a self, in the sense of an
agent controlling overt behavior, then an appropriate method
should detect it invariably. The appropriate method for testing
whether a given condition is under control consists of disturbing
the presumably controlled condition and noting whether it is
restored to its previous state. The self-concept or self-image
most certainly would be under control, if the self is indeed a
control system. Hence, our basic method consisted of attempts to
find a way of invariably disturbing the self-concept.

     We now report on three programs of experimentation, with
replications, intended to serve two purposes: first, to test the
self as a control system, and second, to refine the experimental
conditions needed to detect its presumably invariable function-

                          First Design

     Do subjects correct false attributions of trait categories?
The first series of experiments was based upon traditional meth-
odolgy similar to the design used (for example) by Frey and
Stahlberg (1986). We reasoned that if subjects were presented
with descriptions of themselves that were partially false (that
is, contradicting their own self-attributions), actions to cor-
rect the incorrect attributions would demonstrate the existence
of an agent regulating the self-concept.

     We used the Personality Profile of Conte and Plutchik (1981)
as the instrument for securing the self-attributions, because it
groups an 89-item adjective checklist into eight factors
presented in a circumplex "pie", giving a simple picture of one's
person-ality. Carver and Scheier (1981, chapters 4-7) cited
studies tending to show that people construct schemas, or
category systems, which are used in dealing with specific condi-
tions which have not been previously encountered. On the basis of
these findings we assumed that subjects would control the "pic-
tures" they presented in doing the check list with categories
similar to those of the 8-factor analysis.


     Subjects were students in successive sections of two years
of a freshman mental health course. They rated the Conte and
Plutchik adjective checklist for present self and ideal self
early in the course, and were promised a report on the results a
few weeks later. At the time of the report they were first shown
a blank model of the Conte and Plutchik circumplex and asked to
mark on it the relative positions at which they thought their own
scores would fall on the eight factors -- Sociable, Accepting,
Submissive, Passive, Depressed, Rejecting, Aggressive, Assertive.
They were then given a "Personality Profile Report" (PPR) that
contained statements to the effect that (1) "your most
descriptive dimensions were...," (2) "the dimensions on which you
indicated that most items were unlike you were...," and (3) "the
dimensions you would most like to change were...."

     The experimental manipulation exchanged the middle items of
the top three factors and the bottem three factors and reversed
the middle category of the third statement (which reported the
factors showing the greatest discrepancy between self and ideal)
in the PPR. Thus, if a subject's factor scores had been highest
for Aggressive, Accepting, and Sociable and lowest for Passive,
Submissive, and Depressed, then Accepting and Submissive were
interchanged in the PPR. In addition, reversal of the category
most discrepant between self and ideal meant that, if the sub-
ject's ideal was higher than his or her self in that category,
the report stated that he or she wanted to become less that way.
This "report" was followed with three questions asking the sub-
jects, "What is your reaction to the description of the dimen-
sions (like you/unlike you, etc)?"

     We attempted to prime the subjects' sensitivity to inaccur-
acies in the report by announcing when the materials were handed
out, "These are new instruments and we are not sure whether they
are entirely accurate yet."

     Subjects' reactions to the test questions on the PPR were
scored as follows:

     S corrected a false attribution; for example,"The test is
  wrong, I'm not that passive."

     S corrected a category that had not been manipulated; e. g.,
  "I don't think I'm that aggressive."

     S accepted the report as is; e. g., "It seems O K."

     S's response was unscorable because of being irrelevant
     or self-contradictory.

Results of First Design

     The results of the scoring for the first three data samples
are shown in Table 1.



       Table 1. Reactions to receiving false attributions

Subject N Corrections Acceptances Corrections Ambiguous
Sample of disturbed of false of undisturbed
Number categories attributions attributions

   1 10 9 10 6 5
   2 12 13 9 6 8
   3 12 8 2 18 8
   Note: Each S responded to three questions, hence frequencies
          show number of chances to correct, that is, three times

          the number of subjects.

     It is apparent that the results in Table 1 do not show
strong support for the hypothesis. In sample 1 more subjects
accepted false attributions than attempted to correct them.
Comparing instances of correcting false attributions with
combined instances of accepting false attributions and changing
unmanipulated attributions, we find the number of failures to
correct is 61 as against only 30 instances of correction.

     The problem after examining the results of the first sample
was one frequently faced by investigators, namely that of decid-
ing whether the method used had shown a refutation of the
hypothesis or whether it was inadequate to the task of testing
the hypothesis. Following a long honored tradition in
psychological research, we chose to believe that the method was
inadequate. At this point it occurred to some members of the
research team that perhaps the procedure was too simple. One
member of a study group looking at the results, a school teacher
familiar with students similar to the subjects, speculated that
many of the subjects probably did not have a very strong
self-concept, especially with the kind of trait categories being
used in the design.

     Therefore, we went back to the drawing board, attempting to
isolate a confounding factor such as ego strength or clarity of
the self-image. The data for a potential measure were already at
hand, inasmuch as the subjects' predicted self-profile could be
compared with the actual analysis-program results. The discrep-
ancy between the predicted and actual values of each factor on
the two "pies" was calculated and summed for each S, and a mea-
sure called "self knowledge" was tentatively defined as the
inverse of the total.

     Table 2 presents the results of the cross tabulation of the
correction scores split by the median on self knowledge.

  Table 2. Sample 1 Ss' reactions to receiving false attribu-
               tions, by high and low self-knowledge groups.

         N Corrections Acceptance Corrections Ambiguous
              of false of false of undisturb-
            attribution attribution ed attribution
High 5 | 1 5 7 2
Discrep- |
  ancy |
Low 5 | 8 4 0 3
Discrep- |
Note: High discrepancy is equivalent to "low self knowledge" and

     This time it appeared that the five subjects defined as hav-
ing greater self-knowledge performed considerably more as hypoth-
esized than did those in the other half of the sample. A chi-
square test comparing the first column against the next two lump-
ed together was significant at the 5% level, and so we might have
submitted for publication.

     Instead, however, we made the mistake of attempting to
replicate. See Table 3.

     Table 3. Sample 2 Ss' reactions to receiving false attribu-
               tions, by high and low self-knowledge groups.

         N Corrections Acceptance Corrections Ambiguous
              of false of false of undisturb-
            attribution attribution ed attribution
High 6 | 8 3 2 5
Discrep- |
  ancy |
Low 6 | 6 2 5 5
Discrep- |
  ancy |
Note: High discrepancy is low self knowledge. One high dis-
        crepancy S was scored ambiguous because of loss of data.

     Table 3 supports the null hypothesis; it presents a picture
directly the reverse of Table 2, possibly suggesting that in this
sample Ss with more self-knowledge were less likely to correct
false attributions. Some of the data for the measure of self
knowledge were lacking in sample 3, therefore, the self-ideal
discrepancy from the initial data of this sample was used as a
possible alternative. The results again supported the null
hypothesis. See Table 4.


  Table4. Sample 3 Ss'reactions to receiving false attribu-
            tions, by high and low self-knowlege groups.

         N Corrections Acceptance Corrections Ambiguous
              of false of false of undisturb-
            attribution attribution ed attribution
High 6 | 5 2 8 3
Discrep- |
  ancy |
Low 6 | 3 0 10 5
Discrep- |
  ancy |

     At this point the initial hypothesis was properly abandoned.
In the course of gathering these data, however, incidental
remarks made by the subjects during the debriefing discussions
suggested a closer examination of the theory.

     For one thing, comments gathered from the subjects by the
research assistants, who obtained the data in samples 2 and 3
suggested that many subjects' intentions in completing the reac-
tion sheets were considerably more complex than we could have
wished them to be. Several subjects stated that the manipulated
results didn't seem quite right to them, but they had neverthe-
less accepted them, for various reasons. For example, there were
a number of remarks like, "I thought maybe your tests got things
I had wrong about myself." Others stated that they thought
perhaps the tests were picking up changes which had occurred in
themselves during the class since the initial testing.

     As clinicians we might be skeptical especially of the first
type of comment. These remarks, however, did serve to tell us
that we could not be sure that the same self-perceptions being
controlled at the time of the initial data gathering were still
being controlled at the time of reporting. Hence we might have
failed truly to disturb the self-images of many subjects. To
check this hunch, we came up with the second design.

                          Second Design

     This design called for a procedure in which the self-image
disturbance and correction, if any, would be made in the immediate
situation. One of us (D G) had been experimenting with clinical
applications, feeding back Q-sort self-images, and had got intriguing
results. In our second design we had our subjects create a Q-sort
self-image by selecting from the Conte and Plutchik list 16 adjectives
that they confidently felt to be self-descriptive.

     These self-images were then disturbed in the following manner.
Students were asked to pair up to discuss their self-images, one as
"experimenter" and the other as subject. The "experimenters" were then
instructed to begin examining their partner's Q-sort, to read the
most-like-me item aloud, and then immediately to say, "No, you're not...
(whatever the item)," and then to write down the next statement the
partner made. The recorded statements were then scored as in design 1.
That is, if a S said anything like, "I am so," or "You're nuts," it was
called a correction, otherwise a non-correction. Table 5 presents the
results in four samples of subjects using this design.

Table 5. Subjects' replies to contradiction of self-image in
            four samples using the second design.
Sample 1 Sample 2 (N=8)


Correction Non-cor- Ambig- Correction Non-cor- Ambig-
            rection uous rection uous
     8 - - 7 - 1
________________________________ _____________________________
Sample 3 Sample 4 (N=10)


Correction Non-cor- Ambig- Correction Non-cor- Ambig-
            rection uous rection uous
     8 1 1 9 - -
________________________________ _____________________________

Results of Second Design

     These results seem to support the hypothesis that the person acts
immediately and strongly to preserve the self-image he has just previously
constructed in filling out the Q-sort. To say that the person does it is
tantamount to saying that it originates within the self-system of the
individual within Powers's (1973) model, where the self-image is considered
to be the perception controlled by the self-system.

     While this new design appeared to limit the field in which the
subject's purposes could stray from the phenomena which we, the
experimenters, wished to have under consideration, there were other sources
for misgivings about how well extraneous influences were controlled in this
experiment. An obvious one concerned possible hidden flaws in the
deceptive form of experimental manipulation. We have had misgivings about
the possibility of conscious or unconscious collusion between classmates
acting as subjects and experimenters -- telling the investigators what they
wanted to hear. We are skeptical of the descriptions of success of
experimental manipulation in many current deception designs in social
psychology. We therefore went on to the Third Design to come upon the
problem from still another angle.

                          Third Design

     In this design we preserved the condition of self-description and
disturbance in the immediate situation, but eliminated the deception-based
manipulation. A straightforward self-report design satisfied these
conditions. We asked subjects, in a printed questionnaire format, to
choose the 20 "most relevant" items from the full 89-item adjective
checklist and arrange them in a Q-sort from "most like me" to "most unlike
me." Then they were requested to make up five comparable statements about
them selves, arrange them similarly, and then respond to the following:
Imagine that someone whom you know well were to tell you, for each
statment, "No, you are not like that." What would you reply to him or her?

Next they were given this instruction: "Now go back to the list of
adjectives and pick out five which you did not use, in other words, 5
adjectives which are neutral for you, neither like nor unlike you. Write
them below and after each write what you would reply if someone whom you
know well said (for each adjective), "No, that is not like you."

     Ss' replies were scored as in the previous programs. Answers like, "I
am ..(term repeated).." or "Go to hell" were scored as corrections.
Statements like "I know I'm not" or "You're right" were scored as
non-corrections. An intermediate category (modification) was added for
statements where the S modified his or her own description in the face of
the hypothetical challenge by rewriting their self-descriptive item. For
example, if the subject's initial item was, "I am a quiet person," then, if
he or she wrote, "I am quiet among strangers and I don't assert myself in
conversations" (in reply to the hypothetical challenge) it was scored as a
modification. Another example was, "I'm not like that around people I
know, but I'm am (sic) around others." (You can see that examples such as
the above might be scored as corrections with a liberal scoring policy, but
we chose to stick to a more conservative approach in scoring.) Similarly,
we scored a neutral item as modified when, for example, a S who had
initially written "Depressed" wrote in reply to the hypothetical challenge,
"I believe everybody gets depressed once in a while."

Results of Third Design

     Table 6 shows the results for the eight Ss so far employed in the
third Design.

Table 6. Written responses to hypothetical disturbance of self-
           chosen self-descriptive and neutral attributions.


    Reaction to Disturbance Reactions to Disturbance
     of Relevant Statements of Neutral Adjectives

Sub- Correction Modifi- Accept- Correction Modifi- Accept-
ject cation ance cation ance
  1 1 4 0 0 4 1

  2 3 2 0 0 2 3

  3 5 0 0 0 3 2

  4 0 5 0 2 2 1

  5 5 0 0 2 3 0

  6 4 1 0 0 0 5

  7 4 0 1 0 2 3

  8 5 0 0 0 4 1
         --- --- --- --- --- ---
total 27 12 1 4 20 16

     While none of the methods brought perfect results, it seems clear that
most subjects in the second design showed strong tendencies to correct real
(though acted) contradictions of self-attributions, and subjects in the
third design corrected hypothetical contradictions of self-attributions,
immediately following the making of these self-attributions. This is in
sharp contrast to the situation of the First Design experiments, in which a

period of several weeks intervened between making the self attributions and
the feedback of the manipulated reports.

     Our results appear consistent with our prediction from Powers's
theory, namely, that the self would function as a control system,
maintaining the perceptual variable as currently perceived against outside
disturbance. The methodology under development, primitive as it is in its
current state, is moving in a direction which Pryor (1980) noted that Bem
and Allen (1974) and Allport (1937) had implicitly proposed. As Pryor
stated it, "self-reports are related to the behavioral dispositions they
reflect when they are interpreted from an idiographic
idiographic approach analyzes traits in an individualistic fashion." We
suggest that his discussion, like the others which he cites, is
unnecessarily intricate because of the avoidance of saying straight out
that one can make the best sense of another's actions (including those of
subjects in psychological experiments) if one takes the actor's purposes
at the moment into account.

     Just as Robertson and Glines (1985) found that subjects in a complex
learning task on a computer stayed with it long, or abandoned it quickly,
or locked into sterile repetitions going no-where, or engaged in random
observations to gather data -- according to whether they viewed themselves
as good or bad at computer games, or found the tasks interesting or boring
(another variable dependent upon one's self conception), so do we conclude

that the subjects in the present experiments would vary in their actions
according to whether, as individuals, they felt they could learn something
about themselves, please or frustrate the investigators, or whatever other
purposes of their own which might be served in the situation.

     As against the generally low-magnitude correlations that many
social-psychological, and personality studies show when attempting to boil
idographic variables into nomothetic categories, we believe we may be on
the track of a true nomothetic variable, a purpose common to all humans:
preservation of the perceived qualities of self when disturbed by an
external source.

1) Paper originally delivered to Control Systems Group Conference,
Kenosha, Wis. 1987. Edited and slightly revised as to wording by R. J.
Robertson, 1995 and 1997.

2) We wish to thank Prof. Phil Runkel for helpful comments on the original
version of this study.
Sorry it took so long. I have not been able to upload with my
Delrina WinCom program (it downloads OK) since I upgraded the
harddrive and the idiots sold my old one before transferring data.
Something about the re-install maybe. Anyone have any suggestions, I
would appreciate them.
Best, Dick Robertson