EVIDENCE FOR DİSCRİMİNATİON
The results demonstrated strong differentiation on the part of the subjects in that they gave different responses to the balanced and to the unbalanced triads even though ali had negative P /O relations. This finding clearly favors Heider över Newcomb. How-ever, neither Heider nor Newcomb vvould have predicted the result that balanced triads are judged as less pleasant than unbalanced triads —as has been the expectation in this study.
Consistency Ratings
When subjects made judgments of consistency, evidence for discrimination vvas obtained by comparing the consistency ratings of a story, with the consistency ratings of the' other story vvith which it had been paired. Matched (or paired) t-tests which vvere computed separately for each of the 7 treatments, show that (Table 4.1) subjects gave higher consistency ratings to the balanced sto-ries than they did to the unbalanced stosto-ries, even though both stories within a certain pair do belong to Newcomb's "nonba-lance" category. A more detailed version of the same table broken down story by story is given in Appendix B (Table B.l, p. 107).
As can be seen from Table 4-1, subjects vvere able to make this differentiation in the direction expected by Heiderian balance predictions in every single comparison. The differences were large and persistent, and ali of these were sigıı-ficant at less than the
•0005 level (2-tailed), except two comparisons in the maximum / consistency condition which were significant at less than the .001 level.
55
TABLE .1
Means, Standard Deviations, and Matched (Paired) T-test Results in terms of CONSİSTENCY (0=Low, 10=High) Ratings, över ali Stories for each Treatment
FOUR
FOUR UNBALAN-
SIGNIFI-TREATMENT STATISTIC BALANCED CED CANCE
STORİES STORİES
STANDARD Mean 8.50 4.15 .0005
SD 1 .77 2.80
N 30 29
P-IDENTIFICATION Mean 8.24 3.93 .0005
SD 1 .86 2.48
N 30 30
ALL-SENTİMENT Mean 8.39 4.58 .0006
SD 1 .67 2.53
N 30 30
PRİMACY OF Mean 8.78 2.77 .0005
P/O SD 1 .86 1 .71
P/O
N 29 30
DISTRACTION Mean 8.33 4.53 .0005
FROM VVHOLE SD 1 .56 2.96
N 30 29
CONCENTRATION Mean 8.51 3 .89 .0005
ON VVHOLE SD 1 .42 2.28
N 28 29
MAXIMIZING Mean 8.07 2.80 .0005
CONSİSTENCY SD 1 .50 2.17
N 30 30
ÖVER ALL Mean 8.40 3.80 .0005
TREATMENTS SD 1 .66 2.51
N 205f 205f
t Please refer to page 52 for the handling of the missing cases.
Pleasantness Ratings
When the subjects rated the stories in terms of pleasantness, three main findings emerged: (1) Contrary to the pattem observed with the consistency ratings, the subjects rated the unbalanced story of each story pair (excepting one pair in a single condition) higher in pleasantness. (2) Even though the subjects vvere making pleasantness judgments, they vvere able to perceive and report a lot of significant differences betvveen the balanced and the unbalanced stories (Table 4.2) — contrary to vvhat vvould be expected from a Nevvcombian point of vievv. (3) Nevertheless, the discriminations vvere not as large here, vvhen subjects vvere using pleasantness ratings, as they vvere in the case of consistency ratings. As could be
TABLE .
Means, Standard Deviations, and Matched (Paired) T-test Results in Terms of PLEASANTNESS Ratings ( 0 = Low, 10=High) Över AH Stories for each Treatment.
F O U R
FOUR UNBALAN- SIGNIFI-TREATMENT STATıSTıC BALANCED CED CANCE
STORIES STORIES
STANDARD Mean 3 .80 4 . 8 2 .094
SD 1 .98 2.04
N 30 30
P-ıDENTıFıCATıON Mean 2.97 5.53 .0005
SD 1 .74 2.05
N 30 30
ALL-SENTıMENT Mean 3.33 4.61 .008 SD 1 .65 1 .61
N 30 30
PRıMACY OF P /O Mean 3.25 6.00 .0005 SD 1 .77 1 .88
N 30 30
DıSTRACTıON Mean 3.48 5.44 .0005 FROM WHOLE SD 1 .51 1 .71
N 29 30
CONCENTRATıON Mean 2 . 9 2 5.42 .0005
ON WHOLE SD 1 .60 1 .60
N 30 30
MAX1MıZıNG Mean 3.13 6 . 1 2 .0005 PLEASANTNESS SD 2.03 1 .79
N 30 29
ÖVER ALL Mean 3.27 5.42 .0005
TREATMENTS SD 1 .76 1 .87
N 208F 2081
t Please refer to page 52 for the handling of the missing cases.
seen in the detailed table presented in Appendix B (Table B.2, p. 108) some of the story by story eomparisons could only reach relatively moderate levels of significance while some others failed to attain any significance.
I
The results obtained with the dichotomous choices closely parallel those obtained with the ratings. Dichotomous choice data are reported in Appendix C (Tables C. 1 and C. 2).
57
Summary Coacerning Ratings of Consistency and Pleasantness Subjects judged the balanced triads as being highly consis-tent, while another independent group of subjects perceived the same balanced stories as more unpleasant than the unbalanced ones
—ali in üne vvith the expectations of this study. This particular finding is rather important in that. unlike the above mentioned expectation of this study, previous researchers were inclined to assume that Heiderian unbalance implies unpleasant'ness as well as incon-sistency. The above finding (i.e., the tendency for the balanced triads to be rated as polar opposites on the scales of consistency and pleasantness) runs counter to this popular assumption. And there exists the possibility that such assumptions may have played a role in the mixed results obtained by some previous tests of the balance theory.
Even though some results similar to those of this study (i.e., significant pleasantness difference between balance and unbalance, vvith negative P /O bond) occasionally did turn up u;ıexpectedly in some experiments, because they vvere running counter to the expectations at the time, the researchers did not have an ansvver ready at hand to accound for them. Furthermore, previous research materials possibly involve unvvarranted unit relations vvhose implications vvere discussed in depth in the revievv of the literatüre.
In sum, consistency judgments partitioned the nonbalanced triads in the vvay predicted by Heider's theory. The finding that pleasantness judgements, too, even though to a lesser extent, did display a tvvo category partition runs counter to Nevvcombian pre-dictions. Hovvever the fact that pleasantness judgments shovv a pattern of results opposite of that vvhich is observed vvith consis-tency judgments runs counter to both Nevvcomb and Heider. This vvas explicitly predicted by this study. Further, the results support-ing this expectation (together vvith the that discrimination based on pleasantness judgments are not as povverful compared to consis-tency judgments) serve as evidence for my conclusion that the pleas-antness rating task is an inappropriate response index to be used for testing balance predictions.
58
COMPARISON OF RESPONSE MEASURES (CONSİSTENCY VERSUS PLEASANTNESS)
Ratings
Here, the comparison of interest is between the nature and the magnitude of the differentiation made in terms of consistency ratings and the one made in terms of pleasantness tatings. The pur-pose of the comparison is to see whether the t\vo types of dependent measures would enable the subjects to make the same type and level of discrimination, or whether the two would behave somewhat differently. For this purpose, mean differences of ratings between bal-anced and unbalbal-anced triads (MDR's) are computed: The MDR's for consistency are easily computed from the bottom row of Table 4.1. The MDR's for pleasantness are similarly computable from the bottom row of Table.4.2.
TABLE 4.3
Mean Differences of Ratings (MDR's) of Balanced Triads Minus Unbalanced Triads, vvith Consistency and Pleasantness, across Trearments
Configuration Pairs Response t 1
Compared Measure MDR P
Seascape Mask CONSİSTENCY 4.45
( - + - ) ( - + + ) .0005
PLEASANTNESS -2.63 Soap Opera Jukebox CONSİSTENCY 4.68
( +) ( - + + ) .0005
PLEASANTNESS -1 .53
Wagner 'Dali CONSİSTENCY 4.67
( + ) ( — ; — - ) .0005
PLEASANTNESS -1 .92
River Swamp CONSISTENCY 4.67
( - + - ) ( - ) .0005
PLEASANTNESS -2.49 Över ali Stories CONSİSTENCY 4.60
.0005 PLEASANTNESS -2.17
t Based on t-test of the MDR consistency Vs MDR pleasantness; in other vvords, a test of the difference betvveen differences.
We can see from Table 4.3 that the results for the two depen-dent measures show some striking differences. In the last celi, 4.60 and -2.17 stand for the difference scores (summed and averaged across ali treatments and stories) between the ratings given to the
59
balanced stories and to the unbalanced stories, in terms of consis-tency and of pleasantness, respectively. To look at those difference scores more closely, we can use the follovving representation (Figüre 4.1) showing grand means and mean differences in ratings.
Figüre 4.1
SCHEMATIC REPRESENTATİON OF LOCATION OF GRAND MEANS FOR BALANCED AND UNBALANCED TRİADS, CONSİSTENCY AND
PLEASANTNESS
UNBAL. G.M. BAL. G.M.
INCONSISTENT
k.60»MDR
3.79 .1+1 10 CONSISTENT
U N P L E A S A N T
HAL. G.M.
PLEASANT
UNBAL. G.M.
We note here that the difference scores for consistency and pleasantness are of opposite signs, vvhich indicates that the kind of differentiation betvveen balanced and unbalanced triads (given negative P /O) in the pleasantness case is the opposite of that in the consistency case. In other vvords, given negative P /O, a balanced triad is perceived as more unpleasant than an unbalanced triad, vvhereas an unbalanced triad is perceived as more inconsistent than a balanced triad. I refer to this as the difference in the predictive nature of consistency versus pleasantness.
Given that the balanced ratings get a high rating on the consis-tency scale, while the same stories get a low rating on the pleasant-ness scale, the difference of the ratings, calculated as a subtraction of unbalanced ratings from balanced ratings, gives a positive rating difference in terms of consistency vvhile it results in â smaller but negative rating difference in terms of pleasantness. The above fact represents one of the most important findings in the study: that pleasantness judgments differ from consistency judgments both in kind and in the magnitude of the discrimination subjects make vvhen comparing the balanced and the unbalanced triads. When dealing vvith triads having a negative attraction relation, pleasantness acts only as a negative predictor of "balancedness"—i.e., the balanced stories (given - P /O) vvould be perceived as more un-pleasant than the unbalanced ones.
60
Another vvay of looking at these results is by examining the sheer amount of discrimination that each of the two response measures permits betvveen balanced and unbalanced triads. For this purpose, vve can, for a moment, ignore the role the sign of the difference scores plays, and concentrate on their sizes alone. We can do this by comparing the absolute values of mean differences as sum-marized in Table 4.3 and 4.4 presenting the results treatment by treatment and across stories.
TABLE 4.4
Absolute Values of Mean Differences of Ratings (|MDR|) of Balanced Triads Minus Unbalanced Triads, using Consistency and Pleasantness, across Stories, for each Treatment
Treatment Consistency
Ratings
Pleasantness
Ratings P
STANDARD 4.35 1 .01 .0005
P-IDENTIFICATION 4.31 2.55 .001
ALL-SENTIMENT 3 .81 1 .28 .001
PRİMACY OF P/O 6.01 2.75 .0005
DISTRACTION FROM THE VVHOLE
3 .80 2.01 .018
CONCENTRATION ON THE VVHOLE
4.61 2.50 .007
MAXIMIZING 5.26 3.09 .005
It is clear that, even in terms of the sheer amount of discrim-ination vvhile ignoring the reversal of signs, the pleasantness meas-ure does not function as vvell as consistency does; subjects' ratings of pleasantness do not differentiate the balanced from the unbalanced stories to the extent that their ratings of > consistency do. In the story by story comparisons, hovvever, as presented in Table 4.5, vve can see that, vvith certain story pairs, pleasantness can produce absolute differences as large as consistency does.
The drastic differences in the pattern of results obtained for pleasantness and for consistency vvhen testing balance prin-ciples, can be shovvn by means of Table 4.5 presenting the
compar-61
TABLE 4.5
Comparison of Response Measures in terms of Mean Differences in Ratings (MDR) of Balanced Minus Unbalanced Triads:
(P-values shown are for Consistency MDR Vs. Pleasantness MDR; Group T-tests)
TREATMENTS
•
RM*
(Seascape -Mask)
(Soap Opera Jukebox)
- (Wagner-Dali) (RiVer-Svvamp)
Över ali (Balanced -Unbalanced)
MDR P MDR P MDR P MDR P MDR P
STANDARD CST
PLE
3 .67 -2.47 *
5.00 0.47 *
3 .76
-1.17 * 5 .40 -0.90 *
4.35 1 .02 *
P-1DENTIFICATION
CST PLE
5.00 -3 .10 *
4.60 -2.20 *
3 .97 -2.30 *
3 .70 -2.63 *
4.32 -2.56 * ALL-SENTIMENT
CST PLE
3.60 -0.20 **
3.57 -0.83 *
3 .93 -1 .57 *
4.17 -2.53 *
3 .82 -1 .28 *
PRIMACY OF P/O
CST PLE
6.50 -3.03 *
5.38 -1 ,83 *
6.60 -2.33 *
5.63
-3.80 * 6.02 -2.75 *
DISTRACTION CST 3 .60 3.53 3 .86 4.60 3 .80
FROM THE WHOLE PLE -3 .07 * -1 .38 * -1 .17 * -2.23 * -2.00 *
CONCENTRATION CST 4.80 4.62 4.41 4.27 4.61
ON THE WHOLE PLE -4.33 * -2.03 * -1 .87 * -1 .80 * -2.51 * MAXIMIZING
CST PLE
3.97 -2.23 *
6.10 -2.90 *
6,10 -3 .10 *
4.90 -3 .50 *
5.27 -3 .09 *
ÖVER ALL CST 4.45 4.68 4.67 4.67 4.60
TREATMENTS PLE -2.63 * -1 .53 * -1 .92 * -2.49 * -2.17 »
isons in full detail. The t-test comparisons of consistency mean difference ratings and pleasantness mean difference ratings gave strong, highly significant results. The reader may find it helpful to note that the bottom row of Table 4.5 is the same as Table 4.3, and that the MDR's of Table 4.5 are derived from the data in Tables 4.1 and 4.2.
Dichotomous Choices
Table 4.6 showing the "polar opposites" nature in the accu-mulation of the dichotomous choices further substantiates our find-ings. Chi-square results are impressive; the chi-square values for each pair of stories are ali significant ( .0005). A more detailed table sho A'ing also the treatment by treatment breakdown of the results is presented in Appendix C (Table C. 3, p. 114).
With the help of Table 4.7 and 4.8 which present the overall results of the dichotomous choices, the majör point of the results becomes visible to the naked eye. We conclude from ali the above that the judgment of pleasantness is clearly a "different animal"
from the judgment of consistency.
Pleasantness produces not only opposite results, but smaller and more variable discriminations. Since subjects do not make as much differentiation betvveen balanced and unbalanced stories vvhen judging pleasantness, the failure to find differences betvveen
Heider-iaıı balance and unbalance (vvhen both are of the Nevvcombian
"nonbalance" type) can be attributed to the use of the inappro-priate dependent measure.
Summary of Results concerning Comparisons of Heider versus Newcomb
The results of the tests of the relevant hypotheses that were presented in Chapter 2 are summarized belovv: Hypothesis 1 (p. 32) is confirmed as can be seen in Table 4.1 (p. 56). Similar results can be seen in Appendix B (Table B. 1) in terms of story by story break-dovvn, and in Appendix C (Table C. 1) in terms of the dichoto-mous choices of inconsistency.
Hypothesis 2 (p. 32): In the case of the dichotomous choices, this hypothesis was not totally confirmed but only in the techni-cal sense. In the standard /pleasantness condition, the Z-square value for one story pair came out to be significant while a non-significant difference vvas predicted for ali pairs. Hovvever, as the inspection of Table C.2 in Appendix C reveals, the significance
63
TABLE 4.6
Number of Subjects Choosing Balanced or Unbalancest Stories as Inconsistent, and Number Choosing the Same Stories as Unpleasant, Över ali Treatments.
CONFIGURATION PAİR P/O
P/X
o / x CONSİSTENCY PLEASANTNESS
Z'
SIGNIFICANCE
SEASCAPE + 21 151
AFRICAN
MASK + + 184 59
159.9
\
.00005
SOAP OPERA
+ 17 120
106.1 .00005
JUKEBOX + + 186 .93
WAGNER
+ 16 159
189 .6 .00005 189 .6 .00005
DALI — 188 53
RIVER + 22 184
163.3 .00005 163.3 .00005
SWAMP — 154 57
64
TABLE .
Discrimination via Dichotomous Choicesf (combined across Treatments) BALANCED UNBALANCED Inconsistency
Choices 76 742 818
Unpleasantness 584 262 846
Choices
660 1004
t Shovvs total number of "votes" obtained by stories vvith the corresponding configuration.
TABLE 4.8
Discrimination via Dichotomous Choices (combined across Treatments)
In terms of rovv percentages:
BALANCED UNBALANCED Inconsistency
Choices Unpleasantness Choices
9.3 % 90.7 %
69.0 % 31.0 %
100 %
100 %
level for this particular pair in this condition is only at the lowest conventional level, while those of the other conditions are better than that in this condition. In the case of the ratings (Table 4.2), this hypothesis was confirmed for the standard /pleasantness con-dition again for the 3 pairs of stories out of a total of 4 pairs.
Please see Table B.2 in Appendix B for a story by story breakdown of the results (p. 108).
Hypothesis 3 (p. 32) is discussed, together with the other hypotheses about variation, at a later section (see page 76).
Hypothesis 4 (p. 32) was not confirmed, except in the case of the comparison across ali stories. Even though this hypothesis in its present form was not confirmed, this failure can also be interpreted as only a technical shortcoming, because the intent of the hypotheses need not be limited to their present wording. That is to say, a significant negative correlation (as is the case here), although an insignificant correlation was expected, can also be used as an indi-cator of the idea that the pleasantness ratings do not necessarily measure the same thing that the consistency ratings do.
In this section as a whole, the results of the above hypotheses serve to support Heider's position against Nevvcomb.
65
CHAPTER 5