CHAPTER 5
TABLE 5.1
Analysis of Variancet of Composite J Consistency Ratings for Treatment and Configuration (Balanced /Unbalanced) Effects
Source SS DF MS F P R2 ı
Main Effects 2286.6 7 326.7 75 .0 .001 .56 Treatment 49.8 6 8.3 1 .9 .078
Config. 2236.8 1 2236 .8 513.5 .001 Treat. X Config. 56.0 6 9.3 2.1 .047 Explained 2342.6 13 180.2 41 .4 .001 Residual 1768 .7 406 4.4
Total 4111.3 419 9.8
t See page 52 for note on missing values.
X Composite rating for balanced (unbalanced) triads is the average rating of 4 balanced (unbalanced) stories, for each subject.
TABLE 5.2
Analysis of Variance f of Composite X Pleasantness Ratings for Treatment and Configuration (Balanced /Unbalanced) Effects
Source SS DF MS F P R2
Main Effects 501 .4 7 71 .6 22.2 .001 .27 Treatment 20.3 6 3.4 1.1 .392
Config. 481 .1 1 481 .1 149.1 .001 Treat. X Config. 50.0 6 8.3 2.6 .018 Explained 551.4 13 42.4 13.2 .001 Residual 1309.2 406 3.2
Total 1860.6 419 4.4
t See page 52 for note on missing values.
X Composite rating for balanced (unbalanced) triads is the average rating of 4 balanced (unbalanced) stories, for each subject.
sistency ratings display a main effect (p < .001) due to the Configuration factor, but there is no main effect of Treatments.
However, the two-way interaction of Treatmeıt with Configuration is significant ( p < 047). The results in the case of pleasant-ness are very similar to those of consistency.
A 3-way ANOVA (Table 5.3) vvas run to see the contribution, also of the Types of Rating (consistency /pleasantness) relative to the contribution of Configuration and of Treatments, tovvard ex-plaining the total variation. Again, the Treatments factor does not have a main effect; but the other tvvo do have main effects (Configuration: p < - 0 0 1 ; Type of Rating: pc.OOl).
There is a tvvo-vvay interaction betvveen Type of Rating and Configuration (p < .001). Further, the Treatments factor shovvs
TABLE 5.
Analysis of Variancej of Composite J Ratings for Type of Rating (Consistency v.s.
Pleasantness), Treatment and Configuration (Balance/Unbalance) Effects
Source SS DF MS F P R1
Main Effects 971.4 8 121 .4 32.0 .001 .15 Rating-type (R) 637.5 1 637.5 168 .2 .001
Treatment (T) 12.3 6 2.1 0.5 .999 Configuration (C) 321.6 1 321.6 84.9 .001 2-way Interactions 2473 .7 13 190.3 50.2 .001
R x T 57.8 6 9.6 2.5 .019
T x C 19.7 6 3.3 0.9 .999
R x C 2396.2 1 2396.3 632.3 .001
3-way Interactions \
R x T x C 86.3 6 14.4 3.8 .001 Explained 3531 .4 27 130.8 34.5 .001
Residual 3077.3 812 812 3.8
Total 6608 .7 839 839 7.9
t For note on handling of missing values see page 52
t Composite rating for balanced (unbalanced) triads is the average rating of 4 balanced (unbalanced) stories, for each subjcet.
a two-way interaction (p<.019) with Type of Rating; but it shows no such interaction vvith Configuration — contrary to the results observed in one-vvay and tvvo-vvay ANOVAS. Hovvever, the Treat-ments factor took part in the three-vvay interaction (of ali three factors) vvhich vvas also significant (p < .001).
We novv can go through the above findings interpreting each in turn. First vve see that the one discrepancy betvveen the re-sults from different level ANOVAs is only apparent, as is shovvn belovv. The interaction of the Treatments factor vvith the Configu-ration factor in the 2-way AVOVAs (vvhich vvas in line vvith the results from the one - vvay ANOVAS) indicate the presence of over-all Treatment effects only in the case of the unbalanced triads.
This can be interpreted to be due to a ceiling effect on the balanced triads: Because the balanced triads happen to be located tovvard the extreme ends on both of the scales (hovvever, on opposite poles);
they cannot vary as much, from treatment to treatment, as the un->
balanced ones can, vvhich are located more tovvard the middle and avvay from the ends of the tvvo scales. Nevertheless, this effect must have been rather mild since it disappeared at the level of the 3-way ANOVA. Even though the 3-way ANOVA failed to shovv a
Treat-69
ment by Configuration interaction, it did show a triple interaction — which seems more meaningful as can be seen in a later discussion.
The 3-way ANOVA results provide us with a bird's eye vievv of the findings from the study. At the outset, we observe a persist-ent main effect for the Configuration factor which telis us that the ratings obtained by the two Configurations (i.e., balanced and unbalanced) are different: That is to say that balance and unbalan-ce are not perunbalan-ceived as similar, as Newcomb vvould expect.
Rather they are perceived as dissimilar entities, by and large re-ceiving different ratings across the board, whether measured via con-sistency or via pleasantness (even though to a lesser extent with pleasantness than with consistency) and also across various treat-ments. This observation is supported by the two-way interaction of Configuration and Type of Rating: Not only is the balanced configuration rated differently from the unbalanced configuration, but their respective patterns also change with type of rating being employed. Thus we see that, the balanced configuration is rated as more consistent than the unbalanced one; but the balanced config-uration is also rated as less pleasant than the unbalanced one.
For a pictorial description of this phenomenon which I have elsewhere referred to as the "polar opposites" nature of the two types of rating tasks, please refer to page 60.
Further, we observe a Type of Rating main effect: There are larger differences with the consistency rating than with the pleas-antness ones, between balanced and unbalanced triads. Further, the consistency measure utilizes both ends of the scale, while the pleas-antness ratings tend to cluster toward the lower half of the scale.
Thus pleasantness ratings produce lower ratings as a whole, since both the balanced and the unbalanced triads are perceived as rath-er unpleasant in nature.
There was no main effect for the Treatments factor. However, a two-way interaction of the Treatments factor with Type of Rating is significant. This telis us that the effect of treatments varies with the type of rating task being used. In fact, the effects of treatments are better manifested in the case of pleasantness rating task than that of consistency.
The dependence of the treatment manipulations on other factors is substantiated by the three-way interaction among Treat-70
ments, Type of Rating, and Configuration, which telis us that the treatment effects depend on the particular combination of Rating Type with Configuration: Forexample, the "all-sentiment-relations"
treatment raises the consistency rating of the unbalanced configura-tion while it Iowers its pleasantness rating, this resulting in less discrimination in either case.
On the whole, the above findings show us merely that the treat-ments in general, did produce an overall effect, but they do not teli us which particular treatments were responsible for the results-In order to obtain this information, some other statistical tech-niques were used.
A Priori Contrasts
The hypotheses in Chapter 2 predicted that the difference which may be observed between balance and unbalance configura-tions wi'l show more with some response measures and treatments.
More specifically, for example, with consistency ratings, the distrac-tion treatment was predicted to do poorly, as compared to the Stand-ard condition, in difTerentiating the balanced from the unbalanced triads. On the other hand, the concentration treatment was expected to do a better job than the Standard. It follows that the concentra-tion treatment will also do better than the distracconcentra-tion treatment.
A similar relationship would hold in the case of the all-sentiment treatment versus the maximum treatment — the latter being more successful than the former in effecting a differentiation of balanced-ness from unbalancedbalanced-ness. The results of such a priori contrasts of the treatments taken two at a time, hovvevsr, did not come ouı statistically significant — except when they involved the maxi-mum condition (or the primacy of P /O) being contrasted with the standard, with the all-sentiment, or with the distraction treatments.
The latter three treatments have already been hypothesized, in Chapter 2, to behave similarly. That is to say that these three were expected to do a relatively poor job of discrimination (of balance from unbalance), while the P-identification, the concentration on-the-whole, and the maximizing treatments were expected to produce better discrimination. A priori contrasts in terms of such clustering of similar treatments produced significant results (Table 5.4); Thus those treatments which were expected to increase subject's differential responding to the balanced and unbalanced
71
triads in fact did so. The tvvo contrasts that appear in Table 5.4 vvere constructed in the follovving manner:
Treatment Maximization Concentration P-identification Primacy of P /O Standard Distraction All-Sentiment
Pattern of VVeights Contrast 1
0.5 0.5
0 0 0
-0.5 -0.5
Pattern of VVeights Contrast 2
0.5 0.5 0.5
0
-0.5 -0.5 -0.5 TABLE 5.4
A Priori Contrasts of Groups of Treatments via ANOVA and T-Probabilities, for Consistency, and Pleasantness, in Terms of the Ratings of Balanced Triads,
Unbal-anced Triads, and Difference Scores.
Contrast 1 Contrast 2
Type of Config- Proba-
Proba-Rating uration T Df bility T Df bility Balance} -0.257 200 ns -0.610 200 ns Consistency Unbalance} -0.736 200 .007 -2.435 200 .016
Differencef 2.029 198 .044 1 .632 198 ns Balance} 0 .3243 202 ns 0.3936 202 .046 Pleasantness Unbalance} 2.235 202 .026 2.703 202 .007 Differencef -2.217 201 .028 -3 .021 201 .003
$ Composite rating for balanced (or unbalanced) triads as averaged över the ratings of four balanced (or unbalanced) stories for each subject.
t Difference scores correspond to the balance ratings minus the unbalance ratings.
Trend Analysis
A trend represents a proposed rank ordering of the treatments in terms of hovv successful they are expected to be in effecting a differertiation of balance from unbalance. To form a trend, the treatments vvhich are expected to behave similarly vvere given similar vveights vvhich provided the ranks for ordering them. The trend that comes closest to our consistency predictions vvould have vveights assigned in the follovving manner:
Treatment Clusters Weight Distraction == Sentiment = Primacy 1
Standard 2 P-identification=Concentration 3
Maximization 4
Table 5.5
Trend+ Analysis of Ratings of Balanced and Unbalanced Triads, Difference and Absolute Difference Scoıes, p-Values for Consistency and for Pleasantness (n= 207).
STATISTIC BALANCE UNBALANCE DIFFJ ABSDIFFf '
Pearson r -.08 -.11 .05 .06
P ns .097 ns ns
Kendall tau -.07 -.06 .03 .03
P ns ns ns ns
Spearman rho -.09 -.08 .04 .04
ns ns ns ns
S
+ Proposed trend is: Distraction = Sentiment = Primacy = 1; Standard = 2;
Maximization = 4; P-identification = Concentration = 3.
t DIFF refers to balance ratings minus unbalance ratings.
t ABSDIFF refers to the DİFFERENCE score without taking into account its sign.
A trend analysis was run to test this proposed trend against the data. The resalts are presented in Table 5.5. We observe in Table 5.5 that putting the primacy treatment in the lowest ranking clus-ter seems to have rather a diluting consequence on the treatment effects. In the hypotheses presented in Chapter 2, the "primacy of P / O relation" treatment was expected to do poorly in effecting a discrimination. However, it turned out to be a very successful treat-ment. Because of the unexpectedly large magnitude of its effect, this particular treatment has caused the only serious problem in the ordering of the treatments. The following trend analysis (vvhich further substantiates the results from the a priori contrasts) is directed at making sense out of the effects of this troublesome treatment.
Having concluded that primacy is the only treatment with the direction of its effects contrary to our predictions, I tried another
Pearson r -.08 .11 -.13 .06
P ns .096 .055 ns
Kendall tau -.08 .06 -.06 .00
P ns ns ns ns
Spearman rho -.10 .07 -.08 .00
P ns ns ns ns
73
trend analysis this time leaving out this problematic treatment.
When primacy is taken out of consideration (excluded) from the first trend, we in fact observe higher significance levels for the
treatment effects (Table 5. 6).
TABLE 5 .6
Trend+ Analysis (excluding Primacy) of Ratings of Balanced and Unbalaced Triads, Difference and Absolute Difference Scores, p-Values for Consistency
and for Pleasantness (N=205).
STATISTIC BALANCE UNBALANCE DIFFJ ABSDIFFf
Pearson r -.05 -.22 .15 .16
P ns .004 .042 .028
Kendall tau -.04 -.14 .11 .12
P ns .012 .049 .037
Spearman rho -.06 -.19 .15 .15
P ns .011 .047 .037
P
L Pearson r -.10 .20 -.20 .15
E
A P ns .007 .007 .044
S
A Kendall tau -.10 .13 -.13 .07
N
T P .071 .020 .020 ns
N
E Spearman rho -.13 .17 -.17 .09
S
S P .075 .019 .021 ns
+ Proposed Trend is: Distraction=Sentiment=l; Standard=2; P-identification=
Concentration=3; Maximization=4; Primacy=0 to be eliminated.
İ DJFF refers to balance ratings minus unbalance ratings.
t ABSDIFF refers to the DİFFERENCE score without taking into account its sign.
The two trends which were presented above had been con-structed to correspoııd to our predictions associated with the consis-tency ratings. The foIlowing set of trends are now formulated to fit our expectations associated with the pleasantness ratings. In the case of pleasantness, we remember that the standard treat-ment was hypothesized to rank the lowest. Briefly, the two types of trends were expected to have the following patterns:
74
Ttreaments Consistency Trend Pleasantness Trend
Distraction 1 2 All-Sentiment 1 2 Standard . 2 1 Primacy 1 2 P-identification 3 3 Concentration 3 3 Maximizing 4 4
We note in Table 5.7 that this new arraııgement of treatments fits quite well to the pattern expected specifically of the pleasant-ness ratings — even vvhen the primacy treatment is not excluded.
When it is excluded (Table 5.8), the significance levels are improved.
Thus vve can conclude pooling the results from similar treatments produced treatment effects in line vvith the original hypotheses — the primacy treatment being the only exception.
c 0
N S 1 s
T E N C
Y
p L Pearson r -.13 .18 -.20 .09
H
A P .059 .010 .003 ns
S
A Kendall tau -.12 -.11 -.13 .04
N
T P .028 .037 .017 ns
N
E Spearman rho -.15 .15 -.16 .06
S
S P 0.31 0.34 .017 ns
+ Proposed Trend is: Standard = l; Distraction=Sentiment=Primacy=2;
P-identification =Concentration=3; Maximization=4.
J DIFF refers to the balance ratings minus the unbalance ratings.
t ABSDIFF refers to the DİFFERENCE score vvithout taking into account its sign.
TABLE 5.7
Trend+ Analysis of Balance and Unbalance Ratings, Difference and Absolute Difference Scores, p-values for Consistency and for Pleasantness (N=205).
STATIST1C BALANCE UNBALANCE DIFF{ ABSDIFFf
Pearson r -.08 -.14 .06 .07
P ns .048 ns ns
Kendall tau -.08 -.07 .04 .05
P ns ns ns ns
Spearman rho -.10 -.09 .06 .06
P ns ns ı ns ns
75
TABLE 5.
Trend + Analysis (excluding Primacy) of Balance and Unbalance Ratings, Difference and Absolute Difference Scores, p-Values for Consistency and for
Pleasantness (N=176) C
o N S ı
STA TISTIC BALANCE UNBALANCE DIFF X ABSDIFF t C
o N S ı
Pearson r P
- . 0 7 ns
- . 1 8 .014
.11 ns
.12 ns s
T
Kendall tau - . 0 7 - . 1 2 .09 .09 E
N C y
P ns .039 ns ns
E N
C y Spearman rho - . 0 9 - . 1 5 .11 .12
P ns .037 ns ns
P
L Pearson r - . 1 4 .23 - . 2 5 .13
E
A P .054 .002 .001 .080
s
A Kendall tau - . 1 2 .15 - . 1 6 .07 N
T P .029 .006 .005 ns
N
E Spearman rho - . 1 6 .20 - . 2 1 .09 S
s P .031 .005 .006 ns
+ Proposed Trend is: Standard=l; Distraction=Sentiment=2; P-identification = Concentration=3; Maximization=4.
i DIFF refers to the balance ratings minus the unbalance ratings.
t ABSDIFF refers to the DİFFERENCE score vvithout taking into account its sign.
Explained versus Unexplained Variation
The hypotheses concerning variations were originally conceived in terms of variances or "spread" (implying total variation). Hovv-ever, the variation comparisons in terms of the variation explained by the tvvo configuration categories (balance, unbalance) and the residual unexplained variation are more meaningful; and in this form, the comparisons vvould include both those about "ele-vation" and about "spread". Thus, a majority of the hypotheses can be recast into the form of explained and unexplained variation.
Hovvever, the follovving hypotheses have been stated in terms of the unexplained variation in their original form:
Standard conditions, Hypothesis 3 (p. 32):
standard / consistency unexplained variation > sentiment / pleasant-ness unexplained variation.
76
All-sentiment-relations conditions, Hypothesis 3 (p. 32):
all-sentiment / consistency unexplained variation> Standard / con-sistency unexplained variation; and Hypothesis 9 (p. 35): all-sen-timent / pleasantness unexplained variation < standard / pleas-antness unexplained variation.
Primacy of P / O conditions, Hypothesis 12 (p. 37): primacy/
consistency unexplained variation > standard / consistency unex-plained variation; and Hypothesis 13 (p. 37): primacy/pleasantness unexplained variation < standard / pleasantness unexplained vari-ation.
Distraction-from-the-whole condition, Hypothesis 16 (p. 38):
distraction/consistency unexplained variation > standard /consistency unexplained variation.
Tn order to learn more about the specific discrimination phe-nomena associated with the particular treatments, while at the same time testing the above hypotheses, separate one-way ANOVAS (Rating by Configuration) were run for eaclı treatment
—7 for consistency and 7 for pleasantness, altogether 14 one-way ANOVAs. Their results are shown in Tables 5.9, 5.10. Since the results are ali derived from one-way ANOVAs, the within-groups variation can be readily attributed to the variation due to individual differences in responding.
Another table (Table 5.11) can be derived from these ANOVA tables to shovv the proportions of variance (R2) accounted for by the discrimination of balance from unbalance (variation between Balanced and Unbalanced categories). Alternatively, its mirror image, a table shovving the levels of unexplained variation (within the Balanced or Unbalanced categories) associated vvith eaclı treatment could also be construed.
The results of variation hypotheses can be seen in Table 5.11, p.80 . One can see that the results in the table closely parallel the expected pattern of results in the research design (p. 46)—
primacy being the only exception. The same pattern has also been approximated by the order in the trend analyses (see pp. 72-76).
Hypothesis 3 (p. 32) is confirmed. Hhpothesis 8 (p. 35) is tech-nically confirmed, but the decrement is quite negligible. Hypo-thesis 9 (p. 35) is confirmed.
Hypothesis 12 (p. 37) is not confirmed. Hypothesis 13 (p. 37) is confirmed. Hypothesis 16 (p. 38) is confirmed.
77
TABLE 5 .9
Analysest of Variance of Composite+ CONSİSTENCY Ratings for Configuration (Balanced /Unbalanced) Effects
Treat-ment} Source SS df ms F P R!
STCS
Between Within Total
298 .2 312.9 611 .1
1 58 59
298 .2 5.4 10.4
55.3 .001 .49
PICS Betvveen Within Total
279.5 278.7 558 .2
1 58 59
279.5 4.8 9.5
58 .2 .001 .50
SNCS
Between Within Total
218.5 265 .7 484.2
1 58 59
218.5 4.6 8 .2
47.7 .001 .45
PRCS
Between Within Total
547.5 179 .0 726 .5
1 58 59
547.5 3.1 12.3
177 .4 .001 .75
DSCS Between VVithin Total
228 .2 320.0 548 .2
1 58 59
228 .2 5.5 9.3
41 .4 .001 .42
CNCS
Between Within Total
304.9 210.7 515.6
1 58 59
304.9 3.6 8.7
83 .9 .001 .59
MXCS Between Within Total
416.1 201 .7 617.7
1 58 59
416.1 3 .5 10.5
119.7 .001 .66
Över
Ali Between Within Total
2236.8 1874.5 4111 .3
1 418 419
2236.8 4.5 9.8
498 .8 .001 .54
t See page 52 for treatment of missing values.
• Composite rating= S' ratings of balanced stories for b a J a n c e d t j r a d s
for each subject 4
= r a t i n g s o f unbalanced stories for u n b a l a n c c d t r i a d s
4
î CS refers to consistency; ST=Standard; PI=P- identification; SN= all-sentiment relations; PR=primacy; DS=distraction; CN=concentration; MX=maximum.
78
TABLE 5.10
Analyses t of Variance of Composite + Pleasantness Ratings for Configuration (Balanced /Unbalanced) Effects
Treat-ment} Source SS df ms F P R1
Between 15.5 1 15.5 3.8 .052 .06 STPL Within 234.7 58 4.0
Total 250 .2 59 4.2
Betvveen 98 .2 1 98 .2 27 .0 .001 .32 PIPL Within 210.9 58 3 .6
Total 309.1 59 5.2
Betvveen 24.7 1 • 24.7 9.2 .004 .13 SNPL Within 155 .4 58 2.7
Total 180.1 59 3 .1
Betvveen 113 .4 1 113 .4 33 .9 .001 .37 PRPL Within 194.0 58 3.3
Total 307 .4 59 5.2
Betvveen 56.6 1 56.6 21 .9 .001 .27 DSPL Within 150.0 58 2.6
Total 260.6 59 3.5
Betvveen 94.4 1 94 .4 36.7 .001 .39 CNPL VVithin 149.2 58 2.6
Total 243 .6 59 4.1
Betvveen 128.3 1 128 .3 34.6 .001 .37 MXPL VVithin 215.2 58 3 .7
Total 343 .5 59 5.8
Över Betvveen 481 .1 1 481 .1 145 .8 .001 .26 Ali VVithin 1379.5 418 3 .3
Total 1860.6 419 4.4 t See page 52 for treatment of missing values.
\
• Composite Rating = S4, ratings of balanced stories for b a,a n c e d { r i a d s
for each subject 4
= S4, rating of unbalanced stories f u n b a I a n c e d t r i a d s
4
{ PL refers to pleasantness; ST=Standard; PI= P—ıdetification; S N = all-sentiment-relations; PR=primacy; DS=distraction; CN=coricentration; MX=maximum.
79
TABLE 5.11
Comparison of Treatments in terms of Proportions of Variation Explained by Discrimination (Balance v.s. Unbalance Configurations).
TREAT- CONSİSTENCY DİFFERENCE PLEASANTNESS MENTS
(R2) Explained
Var.
Unexplained
in R2
Explained
(cst-ple) (R2) Explained
Var.
Unexplained
STAN .49 .51 .43 .06 .94
P-ID .50 .50 .18 .32 .68
SENT .45 .55 .32 .13 .87
PRİM .75 .25 .38 .37 .63
DIST .42 .58 .15 .27 .73
CONC .59 .41 .20 .39 .61
MX .66 .34 .29 .37 .63
In sum, we can say that the treatments vvhich vvere expected to do a poor job of effecting a differentiation in terms of subjects' responding to the balanced and unbalanced triads, did so — excep-ting the primacy treatment.
The theoretical basis for this expectation (as stated in Chapter 2) has been that insufficient discrimination of the state of balance from that of unbalance (that is to say a lack of Heiderian balance effect) could be due to a failure to cognize the triad as a vvhole.
The distraction-from-the-whole treatment vvas designed vvith this in mind and its respective instnıctions seem to have produced the intended effect (even though not as robust an effect as vvas originally expected).
A similar failure to cognize the triadic situation as a vvhole vvas expected to occur in the case of the pleasantness ratings, because this particular measure seems to have multiple referents each of vvhich is capable of instigating a different response, by diverting the subject's attention to the many different specific aspects inherent in the described situation—vvhich vvould, in the end, result in a lot of individual differences in subject responses.
Moreover, a similar failure (on the part of the subjects) to cognize the triadic situation as a vvhole vvould also be expected in the case of the all-sentiment treatment. The sentimental nature of the descriptive vvording of the situations vvould (a) instigate 80
an affective type of a process which would make subjects' mind tend to linger on concrete things (rather than perform mental operations them to arrive at abstractions); and it would (b) supply a lot of affective material for an affective process to feed on.
Thus, when used together with consistency, the all-sentiment treatment could impair the effectiveness of consistency (in effecting a differentiating response on the part of the subjects) because this treatment would produce a contrary process, interfering with the efforts of the subjects to cognize the wbole situation.
When used with pleasantness, the all-sentiment treatment vvould be vvorking at its prime —inpreventing the differentiation ofbalance from unbalance: Because pleasantness would be seeking multiple and mostly concrete referents among the parts and parcels of a triadic situation; and because the sentimental content of the stories would supply an abundance of such concrete and minute details.
Table 5.11 shows that in every casethe proportion of variation explained by discrimination in the case of pleasantness is consider-ably less than the proportion of variation explained in the case of consistency. That is to say that, the mere substitution of the word
"unpleasant" for the word "inconsistent" in the questionnaires, reduces the explained variation almost by a half. I consider this as evidence for the general cognitive nature of the consistency rating task, and the not-so-cognitive (or "affective") quality of processes typical of the pleasantness rating task, since consistency brings about impressive levels of discrimination.
The presence of affective processes can be detected also in the diverse outcomes associated with some of the treatments when the rating is made in terms of pleasantness. For example, it seems that the treatment of all-sentiment-relations captures the acutest impact of the affective processes: When used together with consis-tency, the all-sentiment treatment lovvers the explained variation even beyond that which the standard condition itself is able to provide (45 % versus 49 %, respectively); this finding is suggestive of the tendeney of affective processes (here instigated by the all-sentiment treatment) to interfere with cognitive ones (as displayed by consistency judgments). On the other hand, as expected, when the all-sentiment treatment is used together with pleasantness,
81
Bqove the level of the standard /pleasantness condition (13%versus 6%, respectively). Yet, the affective quality of the all-sentiment condition, which encourages diversiy in responses, is stili displayed by the fact that, here in the sentiment /pleasantness condition, the level of the explained variation is only less than 1 /3rd of its original level in its consistency counterpart (13% versus 45%, respectively).
it increases, though very little, the explained variation över and There is another treatment, i.e. primacy, which was expected to behave "affectively" by showing differential outcomes when used with pleasantness as opposed to consistency. This differential effect can be observed by studying the same table (Table 5.11).
Yet, this treatment is not altogether similar to the sentiment treatment: On the one hand, primacy is a treatment that is suc-cessful in increasing the explained variation (above the level of the standard) in cases of both consistency and pleasantness. On the other hand, primacy, when used with pleasantness, loses much of this power tö bring about discrimination. This drastic reduction in the amount of explained variation that "primacy" can produce, enables me to categorize primacy together with the sentiment treatment. In fact, (together with the standard treatment), these two treatments undergo the biggest decrements in explained variation This similarity suggests that there is something peculiar about these two treatments so that a switch to pleasantness from consistency hurts thern more than it hurts the other treatments.
To further understand the diverse pattern of sentiment versus primacy effects, results from some other analyses are examined (Table 5.12).
Table 5.12
Comparison of Primacy and Sentiment Treatments
1 Primacy Sentiment
CST PLE CST PLE
Range f 6.1 3.8 4..1 2.4
I Consensus t 29 /30 23/30 26/30 18/30 t "Range" refers to the absolute difference between the ratings given to the
balanced and unbalanced stories.
5 "Consensus" refers to the modal number of dichototomous choices ("votes") a story gets (out of 30 subjects) designating it either as inconsistent or unpleasant.
82
From Table 5.12, it can be seen that the sentiment/pleasantness condition is characterized by very low consensus among subjects as to which story is unpleasant, and by a very narrow range between the ratings of balanced and unbalanced triads. Thus, sentiment represents the circumstances expected of affective process-es—: i.e., an insufficient degree of differentiation and many indivi-dual differences. On the other hand, primacy/pleasantness con-dition shows higher levels (as compared to sentiment / pleasantness condition) of both range and consensus. Even though, this above-mentioned finding was not expected, we also see that primacy / pleasantness condition stili ranks lower, in both aspects, than sen-timent / consistency. Thus primacy approximates sensen-timent in some respects, and there is reason to believe that the two treatments would show a lot more convergence if the variable of structuredness vvere not held constant at a maximal level.
83