Cross-cultural study of game-playing behavior in children : an
interim report
Citation for published version (APA):
Shinotsuka, H. (1975). Cross-cultural study of game-playing behavior in children : an interim report. Technische Hogeschool Eindhoven.
Document status and date: Published: 01/01/1975
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
Dep
75
SRI
CROSS-CULTURAL STUDY OF GAME-PLAYING BEHAVlOR IN
CHILDREN AN INTERIM REPORT
Hirorni Shinotsuka
CROSS-CLJi.TURAL STUDY CF GAHE-PLAY1NG B.EHAVIOR IN
CHILDREN:
AN INTERIM REPORT
Hiromi Shinotsuka* Technische Hogeschool te Eindhoven
1. Introduction.
This is an interim progress report on the cross-cultural study ~roject of which Prof. Toda and I have been the members of the Japaliese team. ïie had done an experiment in Japan in
1969
and ever since have been engaging in the analysis of the whole data obtained in Japan as well as in other countries. ,ie have so far obtained rnany.
interesting results from our data analysis carried out in different levels, from very macro-scopic to very micro-scopic. Nevertheless, as yet we cannot claim that out findings fit together like the pieces of a jigsaw puzzle. So in this report I will select a few of our recent results of analysis, most of which were obtained during my stay in Eindhoven,
duction.
preceded by the general overall results as an
intro-This project was initiated by C.G.McClintock and J.M.Nuttin, Jr. Their ~otivation for the project was to compare the cultural effects upon the development of competitiveness in children. Their major finding was that Belgian boys were much less competitive than Anglo-American boys when the children in the second grade of primary schools were
---.. J:he major part of the work reported here was done under the research fellowship granted ta the author by Technische Hogeschool te Eindhoven. 'l'he .:l.uthor appreciates very much the .lr..ind hela offered b her collea.,.es
t "oU" -" J 0
a ~u, espccially by r~of.Meuwese, Dr.van Vonderen and Mr .Habbinowitsch.
·
compared, but the competitiveness of Belgian boys caught up with American boys at the sixth grade.
In addition to'these two cultures data were collected later from three more cultures, Japanese,Greek and ~iexican-American, and we found that the dyad( the pair of Ss played the game together ) differences were invariably great in any culture. ~o our interest in the data analysis has gradually shifted from the competitive motive to the general empirical model of game-playing behavior. F or example, our current major interest may oe expressed in terms of the following questions j If we are going to categorize the types of game playing
concerning individuals and dyads, what would be the most plausible clasBification? How do these types develop through the course of play? Are there some special events that may trigger a drastic change in the behavior of the players--like the outburst of competitive responses ? Ta sum up, can we predict the Ss' behavior to the later stage of the play fairly precisely by kncwing their early stage behavior? 2. Bxperimental Conditions
Payoff Hatrix: '.ehe payoff matrix in this experiment anà shown in Fig. 1 is a version of Maximizing Difference Game, abbreviate as MD, designed ~o effectively separate the two major motives, generally assumed in experimental game studies, the
motive of maximizins one's own gain and that of maximizing the difference of oue's own gain from the op~onent's. As alternative 'a' dominates
alternative tb' for each player, one cannot increase his own gain by changing his ,"'e.s~~onse from 'a' to 'b'. So i f thè shift occurs, W:re
differ-ence--- which we may simply call the competi.tive motive, and, accordingly ,we may label the respon~e 'bi as the competitive response. We mayalso call the response 'a' as the cooperative response, since the choice of
'a' makes sense, only if the two players both choose 'a', or as au invitaiional reponse intended to lead up to that cooperative state.
Ss: The experiment Ss were chosen from five cultures,
Anglo-America~, JvIexican-American'" , Belgian, Greek and Japanese. They were
boys"'· in the second, fourth and sixth grade of primary schools,as shown in Table 1.
Ap~aratus: Each player used a special response panel on which
were two buttons corresponding to the response 'a' and tb', four feed-back lamps wnich displayed the realized payoff cell when the two
players made tneir responses, and a digital counter showing the sub-total score which he had acquired up to the trial.Under the double display condition explained below, the panel had one more counter to show the op~onentls subtotal score. In addition to these counters the second' graders were also provided with glass tubes into which E ·put'the same number of poker cuips as the number aplJearing in the
counter(s) to help S grasp the meaning of tüe numer{~al ·score(s).
Display Conditions: For each culture and grade, different dyads were run under two display conditions, Single Display and Double
Display, meaning the number of counters. Under the dingle Display
---'" Those who live in California and are usually called Chicanos.
conditivn, there was only one counter for each S, displaying nis own subtotal score. Under the Double Uisplay condition anotner counter was added to suow the opponentls subtotal score. The implication of these two conditions was obvious : Ss were expected to behave more
com~etitivelyunder the Double Dislplay condition which enabled them
"directly cornpare the difference in subtotal scores, thus providing them tue outlet of tue maximizing-tne-difference motivation.
For eücn cümbination of tnree conditions, Culture X Grade X Display, at least 14 àyads were run except with Mexican-Americans. BecauGe of the difficulty in getting enough numger of Chicano subjects, They were run only under the Double Disp~ay_condition.
Experimental ~rocedure: Ss were paired to form dyads,and precautions were taken so that friends did not form a dyad. iach S Séct in front of a panel described above, which obstructed the sight of the opponent. The instruction given to the Ss was written as neutral as possible concerning their motivation, and we didn't advise them
n
to maximize the total gain cotrary to the general practice. We just
l\
told them to play the game. No monetary reward was given.
Each dyad played the game for 110 trials. '.tne last ten trials
wer~ special trials, during the period of whieh experimental assist-ants asked each B On each trial about his intentions, expectatiohs and 50 on. The analysiJof this part of data has not been under the
'.
resp~nsibilityof the Japanese team, and therefore they were omitted
from our analysis.
3.
General Results.The dependent väriable discussed in this section is the relative frequency of competitive responses, and the eïfects of the three experimental väriables, culture, grade and display, together with the trial effect, will be graphically presented.
Fig.2. snows the effect of culture and grade. The competitive-ness increases with grade in every culture. Japanese children are the
most com~etitive, followed by Greek children,Anglo-~ericanchildren
and Belgian children, in this order. The rate of competitive responses of Cnicano children is about the same as that of Anglo-American child-ren for the second graders, but i t does not in~rease with grade as much as the children in the other cultures.
Fig.
3
shows the display X Culture X Trial effect. The solid lines are for the Single Display condition and the dotted lines are for the Double ~isplay condition. The bottom line in the American graph is for the Mexican-Americans who were run only under the Double Display condition. -Chicano children are just about the same in competi-. tivenessas the Belgian children under the sameconditioncompeti-. As trialproceeds, Ss become more competitive in every condition. As expected, children in the Double Display conditivn are more competitive than the children in the Single Display condition.
~vidently, the most conspicuous feature of this graph is
the large difference produced by the Japanese children under the two display conditions. Japanese cllildren under the Single Display condition
show about the same ( or a bit less) competitiveness as the Greek children. Under the Douple Dis~lay condition, however from the first trial block to the second, Japanese children show a large increase in their competitiveness and their rate of competitive response
exceeds 90
%
in the last trial block. In other words, Japanese child-ren are very sensitive to the condition, whether or not they can compare their scores with those of the opponents'. This conforms witn thecommon observation about Japanese culture such that Japanese are other-oriented people and use a relative criteria in evaluating their performance.
This difference in competitivenes~ between the two display
•
conditions is plotted against grade for the four cultures ( excluding Chicanos ) in Fig.4. It is natural that the difference is large with Japanese children, but its decrease with grade should not oe misinter-preted, since i t is mainly due to the fact that with higher grade Japanese children who are quite competitive even under the Single Display condition there is not much room left for the Children under
the Double Display condition to surpass them in competitiveness. The difference in the competitiveness between the two
players forming a dyad (Intra-dyad.difference) will in general indicate the response similarity between the two players. By calculating the distribution of the intradyad differences over all the Ss for each culture, which is obviously symmetric around the zero difference, we obtain the standard deviations for the first and the last 25-trial blocks as shown in table 2.
The standard deviation of Intra-dyad difference within the first 25-trial
bL-:-~k is the smallest with the Japanese Ss and the largest with the Greek
Ss. This mGans that at the beginning Japanese children tend to choose simiiar
resI'cmses to tho<> .. of the opponent.;"f _ "'-no Gree!" children show the least care
about the similarity. Note that Intra-dyad differ~ilce is strictly '~~o~u~tional
to the difference in the accumulated scores between the two players. Because of the strict parallelism between Intra-dyad difference and the score, we may
say that Japanese children do not allow large score differences, while Greek
children apparently do not mind much the score differences. The same tendency persists to the last 25-trial bloek. In comparison to the first trial block results, the standard deviations 1n the last trial block are much higher, indicating that the players tend to accept larger difference between himself and the.opponent.
Let us now seehow the players' response patterns change during the play under eaeh condition. For the purpose of making this elear, we construeted the PPD (standing for Pre-Post differenee) means for eaeh S
in the following way : Aeeording to an ~pplieation of eovariation analysis
teehnique, whieh shall not be diseussed in this paper, it turned out that we
should biseet the whole data around the 25th trial.if we a~e to do it in the
most effeetive way. So, heneeforee, we shall eall the first 25 trials the Pre stage and the remaining 75 trials the Post stage. In order to eompensate the size of the two stages, we ffiultiplied the CR (the number of eompetitive responses) of eaeh player in Pre stage by three, and subtraeted it from his CR in Post stage, to obtain this PPD. Note that the zero PPD means that thc S behaved equally c0mpetitively in bath stages. Likewise positive{negative) PPD means that the S beeame more (less) competitive in Post stage eompareà
.5
dotted lines to Double Display condition. Fig.4 shows that, in general, the,
shift from Single to Double display condition is represented by a displacement of the distribution from left to right, meaning that the additional
information feedback in Double Display condition causes a comparatively uniform
increase in CR in Post stage in each culture, substantiated by the general
covariation analysis indicating a rather small display effect in Pre stage. The rightward displacement is most significant with Japanese children. Also the large variance of the Japanese distribution in Single Display condition
child;~",-indicates that, even though Japanese are on the average the most competitive
11.
among all the cultures studied, there still are those who show a drastic shift toward the cooperative direction in Post stage when the opponent's
score is not explicitly displayed. Fig.6 shows the relation between the
mean and the
sn
of the distributions. Black circles are for Single Displaycondition, white circles for Double Display condition. Standard deviations are all similar except Japanese Single Display condition. About the means, Japanese data exhibit a dramatically large difference between the two display conditions.
L4 . Covariation Analys~~---its principle and the results
In carrying out the data analysis, we needed some analytical tool lo7ith l-]hich we can capture the overallstructure of the whole data ---- a
numerical aid or a compass that will tell us in which Jirection our analysis
·should proceed in the whole complexity of the data structure automatically
suggesting unmanagea~lymany possible alternative courses.
Covariation analysis is a data struçture analysis developed by Toda (Toda,1974) , specifically suited to analyse the kind of data we obtain
of contingency tafules.
In appearallce, the method is somewhatlike analysis of vari.ance applied to contingency tables, since it gives the information in terms of ma1n effects, interactions and residuals. Howéver, it is essentially a non-parametric, llon-statistical technique, providing no significance testing of
its own, even though one can run parallel X2 tests.
The major purpose of this method is to obtain from data the following types of information : By dtcomposing a pooled large data obtained under
e
various different conditions, this method allows us to find which conditions are more important than others, the importance in the sense of the
contribu-tion by the condicontribu-tion to the making of the given data. The method also allo~s
us to estimate which direction of further analysis appears more promising than others.
Although 1 cannot go into details of this method here, let me
just explain the major technical terms. The internal covariation is the
difference between the observed frequency alld the expected frequency calculated
from the two marginal frequencies.(See Fig. 7). The internal covariation
obtained from the pooled matrix is called the total internal covariation w t . The amount of the total internal covariation represents in its own way how much regurarity is there in the pooled matrix, demanding.an explanation in terms of the differential contributions of the conditions. This can be done
by decomposing the pooled matrix according to one or more of the conditions,
and calculating internal covariation w
1' w
z' ..
wi•• for each of the decomposeàsubmatrices. Then the sum!: w. of the internal covariations of these submatrices
1
represents the regularity still left in the data af ter the conditions are
ca11 the difference between the tota1 interna1 covariation and the residual,
h 1 . .
*
t e externa covar1at10n w This external. covariation represents the amount
of the data regura1ity that is explained by the conditions used to decompose the poo1ed matrix. In other words the externa1 covariation"ascribed to each eondition (or each combination of conditions ) represents the amount of
"
contribution of the condition (ol of its combination) to the making of the
who1e data. 1be ma1n effects of and the interactions among the conditions are all defined in terms of externa1 covariations.
Various types of covariation ana1ysis have been app1ied to our data. But 1'11 show here on1y the resu1ts of the covariation ana1ysis concerning the forma11y controlled experimenta1 conditions, with which, for the ease of
interpretation, we random1y selected exact1y 14 àyads for each condition,
and omitted Mexican-American data which 1ack Single Display condition. So in tota1 336 dyads were used in this ana1ysis.
Now let me show the results • The outer circ1e of Fig.8 ShO\>1S the
tota1 interna1 covariation, the main effects of tria1( 4 b10cks of 25 trials) , display, grade and culture, their interactions. Combining them all we can exp1ain 52.6 percent of the tota1 interna1 covariation by these experimenta1 conditions alane. The remainder is the residual. The 1argest main effect is the grade, and the display main effect 1S the smallest.
When we consider the individua1 dyad as acondition variab1e and dëvide the who1e data into 336 submatrices per dyad, the dyad condition exp1ains about 77 percent out of the tota1 interna1 covariation, as shown by the inner circ1e of Fig. 8. This resu1ts ,however, indicate just the size of dyad differences when all the other conditions are neg1ected. Neverthe1ess, it certain1y points out that no satisfactory picture of the data wi11 be obtained un1ess we go much deeper in Dur ana1ysis to the level of individua1
dyads, and this consiàeraticn set our subsequent course of analysis as will
be described in the latter part of this paper. ·Besides that, the results of
covariatioll analysis is quite reasonaole, and with full of important information.
Now let us see the results ~ more closely.
Fig. 9 shows the main display effect and the conditional display
o
effects. In this figure the whole angle of 360 corresponds to the total
external covariation of 759.78 excluding the residual in the outer circle of
-tht
Fig.8. Within~innermost circular band is represented the main display effect
as the three identical sectors. Within the second circular band is represented the three conditional displayeffects ; the left sector being the display
effect given grade, the right sector that given trial Th th· d b d
• _ - . '}I e 1r an
and the bottom sector that given culture.
~----rclpresents the double conditional displayeffects, and the outermost band represents the display effect given the three other conditions. The
discre-pancies between the inner and outer sectors are due to interactio.~~.
There are numerous important messages embedded in this figure, fröm which we shall sample just a fe\-7 ; The amount of display effect is almost monopo-rized by Japanese, especially by Japanese second graders. The display effect
really emerges in the second trial bloik and keeps on. This observation
,
comforms with the graphs shown in Fig. 3,4.
Fig. 10 shows the culture effect. Tbe scale is· same as Fig. 9. From Fig. 10 we can read the followings : Tbe culture effect is greater under Double Display condition than under Single Display condition. There is little cultural differences unàer the second grade Single Display condition. The cultural difference is the greatest with the fourth graders.
Fig. 11 is for the trial effect. Trial effect is small with the
condition. The trial effects for Japanese and Anglo-American are relatively larger than those for Belgian and Greek, but the large trial effect of
Japanese is obtained mostly' under Double Display condition. Fig. 12 1S for
the grade effect. TIle scale unit is changed to the one half of the preceding
figures. The grade effect is larger for Single Display condition than for
Double Display condition, and the smallest in the first trial bloek. It is also larger for Belgian and Greek than for Japanese and American.
5. Markovian Analysis
So far we have examined effects of the formal experimental conditions culture, grade, display and trial by covariation analysis. Though the effects thus revealed are interesting and informative, there still remains a large
residual to be accounted for. In order to improve our descriptive power
of the results, what we must do is to unpool the part of the data whieh
have been pooled,i.e.,the trials (within eaeh trial bloek) and dyads.
However, a mere ~eehanieal dëeomposition of data will not serve the purpose
. What we need to do is to decompose tne data aceording to some struetural criteria eoneerning the game-playing proeesses and dyad charaeteristies. The former leads us to a Markovian analysis, and the latter to a eategoriza-tion of dyads and players. In this seeeategoriza-tion I will briefly describe what
we have done within the context of Markovian types of analyses, temporarily
disregarding differenees in experimental conditions.
Remember that the rules of the game is completely s)~etricalbetween
the two players within a dyad, and two Ss were assigned at rándom as the Row Player and the Column Player. In order to aequire some additional iIlror-mation utilizing this arbitrariness, we shall henceforce discriminate the
two players within a dyad as LC ( less competitive ) and MC ( more competitive ) according to their frequencies of competitive chcices over the whole trials.
When the number of competitive choice was identical between the two players
(though such cases are rare ) , the two players are again assigned as LC
and MC at random. And when the data are pooled, LC and MC are identified
as the row and the column players, respectively. As a result of this
proceèure, the entry of cell 2 is never less than that of cell 3. Note also
that Cell 2 is realizad when LC is cooperative and MC is competitive, and
Cell 3 is realized when the opposite is the case.
"
We then calculated the CR frequencies ~ditional to the cell number
obtained ón the immediately preceding trial, computed over 10-trial blocks.
The result is shown in Fig. 13. Note that at the beginning the CR frequency
is the lowest af ter Cell 4 and is the highest after Cell 1. The two curves
move toward the opposite directions as trial proceeds, intersect at the third
trial block and reach the respective asymptotes,i.e. about .5 for the af
ter-Cell-} curve, and .8 for the after-Cell-4 curve. Although this result is
very interesting, it should not be interpreted fo imply the corresponding direct changes took -place in conditional response probabilities. Note that
in' early trials most of the dyads may have almost equally contributed to
each conditional frequency but on the later stage of play, as each dyad has gradually made its characteristics clear, the more cooperative dyads could have monopolized the After-Cell-} frequencies and the more competitive dyads
the After-CeI1-4 frequencies.
Such is a common pitfall 1n any type of Markovian analysis,
and in order to avoid this misinterpretation it is necessary to take the
characteristics, as we sha11 do in the next section.
Now let us look at the effect of runs of the same ce11s. Fig. 14
shows the CR frequencies p10tted against the 1ength of preceding runs of the same ce11. The re1ative 10cationa1 re1ationships of the four curves are the same as in Fig. 13, though trial b10cks are pooled in this figure. Following the run of 1ength one, meaning that the preceding cell is new1y rea1ized,
the CR frequency does not much depend upon the ce11. ~~en Ce11 1 rea1izes
twice in succesion, however, the CR frequency goes do.m be10w 0.5, and it shoots up af ter two Cell-4s, going beyond 0.9 on the average af ter more than 2 Cell-4s.
Ce11 2 and ce11 3 curves both show neither trial effect (Fig.13)
nor run effect (Fig.14 ). The two curves a1so resemb1e each other as far as these p1ayer-poo1ed representations are concerned. However, a c1ear
difference emerges if one plots the resu1t of the two categories of p1ayers, LC and MC,separate1y, as shown in Fig.IS. What this figure implies are as
fo110ws : As Ce11 2 continue to occur under the sacrificing.act of LC,
MC becomes more competitive as if to exploit the LC. In contrast to this marked tendency, the immobi1ity of After-Ce11-3 curve is significant.
Concerning the after-Ce11-1 and after-Cell-4 curves,note that their
re1ative positions are reversed as the run 1ength increased, the c1ear
message of which is that the continued' rea1ization of Ce11 1 and Ce11 4 shou1d be vita1 to create cooperative and competitive dyads, respective1y. Though not reported in this paper, the idea was persued more deep1y with individua1 dyads, and we now possess a good estimate of the impacts of continued Ce11 1 and Cell 4 over the succeeding choices. In addition to these considerations,
apparent in the first trial block, despite the large trial effect shown in Fig. 13.
6. Categorization of Dyads and Players
As~we have mentioned in the preceding section, it is indispensable to categorize dyads and players to some extent in order to make really
meaningful arguments about the processes of game playing.
It is obviously desirabIe to categorize dyads and players according to some kind of "naturaI" clustering of Ss' behavior characteristics.
CR
Therefore, we have drawn quite -a few scatterograms for
Ol
frequencies ofLC and HC of all the dyads over variou
s
stages (trial-blocks)of the playc~However, the scatterograms were always rather homegeneous, and therefore we have decided to start with a simple mechanical categorization as a start which may later be subjected to modification according to some optimization criterion.
As a matter of fact, the mechanical categorizations we employed
~
worked very weIl, and we do not expect any drastic modification as result ,\
of optimization which is now under way.
The principles of the categorization Version .1 are as follows :
The ineasure used for categorizing players is the CR frequency·in Post-stage.
A player is classified as cooperative, denoted by
Q ,
if his CR in thePost-stage falls below 30. Likewise, a player whose measure falls within the
range of 30 to 45 (inclusive) is classified as neutral, denoted by
! '
andone whose measure is more than 45 is classified as competitive,denoted by ~.
Once players are classified this way, the dyad categorization follows automatically according to the types of LC and MC players of each dyad,
ending up with 6 dyad categories shown in Fig.
lu.
For the sakeof mnemonic convenience we named the ~ dyad categories in the following
way :COOP for
Q
x~Q (standing for cooperators ),. COP.P for ! x ! (standingfor competitors ), .NEUT for,! x N (standing for neutrals ), ~ for
Q
x !(cooperators put against competitors, resulting in a creation of martyrs ).
~ for 0 x N ( cooperators serving neutrals ) and EXPL for N x X
(competi-tors exploit neutrais).
Now let us see whether dyads and players show their categorical characteristics beyond their defining properties, i.e. CR frequencies. Fig.
17 shows the conditional CR probabilities given the preceding cellof LC
and MC in the first and fourth (the last) 25-trial blocks, computed separately for thc six dyad eategories. Note that, beeause of the seareity of samples for Mart dyads, their probability estimates are rather unreliable,
partieularly the one af ter Cell 3. So, even though the Mart results are
interesting, we shall in general disregards Marts in our forthcoming arguments. Now we shall first look at the Pre-stage results. There are rather large non-trivia1 differences among different eategories in the pattern of four points, indieating that, even though the eategory membership is defined on the basis of the Post-stage results, the category eharaeteristies are al ready clear in the Pre-stage. First, pay attent ion to the broken lines
connecting blaek and white triangles,corresponding to After~Cell-2 and
Ater-Cell-3 probabilities. The relative positions of these two triangles are fairly similar in every category (exeept Mart). This is the type of result
we naturally expeet as it is deserib~by the following simple rule : Note
tllat the off-diagonal cells are those whieh give one of the players and extra advantage over the other. One took advantage then will become more cooperative on the next trial, and the one who was taken advantage will beeome more
ccmpetitive ön~ the next trial. Sinee this tendeney is obviously sa natural
e
block, xven though the lengths of the broken lines become much shorter. The relative positions of the two circles on the left hand side
graphs are also similar ~n every category except caMP • So this relation
is also very general in the Pre-stage, which, however, is not at all
preserved in the last stage except for NEUT. With Caap,MART and ca~w,
the direction of toe solid arrows is practically reversed. So apparently some structural changes are taking place through the course of the play.
Now let us get back to the Pre-stage graphs and look at the
NEUT graph. The characteristics of this graph may be very simply described in such a way that each of the two players of a NEUT dyad acts with only two levels of response probability, depending only upon his own previous
response. If his previous response was competitive, his next response ~s
very likely to be cooperative, and vice versa. Remember that a NEUT dyad is formed with two N players (neutrais). The same rule also applies to
N players in other categories. So we may infer that N players are effectively represented by response alternaters ..
The same response alternation tendency is also observed with a and X players who are combined with N players, though their conditional probabilities (given onels own previous response) are of smaller variance and off centered ; those of a are more cooperative and those of X are more competitive.
Then, even though there still are more complications to
be accounted for, particularly about caap, caMP and ~UffiT, we can at least
have a fp.ir hope that a rather simple hypo thesis like the one stated above
may explain a large part of Ssls initial stage behavior.
Now the shift from the left hand side columns to the right of
for response alternating N players' two conditional probability levels is
much reduced. The response alternating tendency for X and
a
playersalmost disappear. The CR probability of X players becomes quite extreme, while that of 0 players do not undergo much change. Partly corresponding to these, only those categories which involve N players do not exhibit the reversal of relative positions of the two circles.
As we have al ready seen of Fig. 14 that, when all data are poole4,
the cell-conditional CR frequencies showed a marked difference for cell-run lengths greater than one, and we had a suspicion that these differences mainly reflected the langer run lengths imposed more selectivity upon the types 01dyadS which could produce such long runs. The suspicion was duely
confirmed ; long Cell runs were produced mainly by
coap
dyads and themajority of long Cell
4
runs were contibuted bycaMP
dyads.So, in order to see the development of dyad category characteristics over trials in a fair, more comparable way, let me select cell-conditional
probabilities given run length and compare t~em among different dyad categories.
Fig. 18 shows these
~un
one cell-conditionals for LC and MC plotted against25-trial blocks, aud categorical differences are apparent even with rUIl 1
,getting more and more enhanced as trial práceeds. The major characteristics
of this figure will be depicted as follows: Af ter Cell 1, the among category
differences are small ön the first trial bloek, but as trial proceeds,
COOP dyads shift toward eooperation(lower left),
caMP
dyads toward eompetition(upper right) and MART dyads go toward lower right, ereating the characteristie MART outeomes.
CO~ dyads invariably develop increasing eompetitive tendeney
,
compared to rather small effects af ter Cells 2 and 3. EXPL dyads show a trend similar to those of CaMP , though the level of competitiveness of the former
is kept lower than the l~tter. MART dyads show the characteristic martyr
shift toward lower right 'irrespecti~~ the preceding cello The curves
for SERV are in general similar to those of COOP ,and ~~UT shows the smallest
trial effect among the six categories.
The implications of all these results are quite complicated,
however, and hardly allow us an unequivocal, unique inteypretation. Itl-order to proceed ahead, we have to build a working model to secure our basic ground and to provide ourselves with a sort of effective information processing system that at least allows us to screen from the tremendous pile of
avail-able information those immediately processavail-able and those calling ~or further
consideration.
Before going into this topic, however, let me just briefly
mention a few other types of results obtained at this stage of our analysis. The change in competitiveness from Pre-stage to Post-stage is plotted in Fig.19 for COOP,COMP,and NEUT categories.(The CR frequencies in the Pre-stage is multiplied by three to enable a fair comparison.) What this graph indicates, among others, is that there are rnany dyads
the members of which change their response propensity rather extremely from Pre-stage to Post-stage, a change that is obviously responsible for the creation of COOP and CaMP dyads.
A detailed Markovian covariation analysis haD been carried out
with cell, trial, category and cell-runs. The overall results are very complicated and I shall not elaborate it here, except.mentioning the
following two points: The rnain effect of the preceding cell is artific.ially
positive external covariation and Cells 2 and 3 also a large negative external covariation, which mutually cancel out. When this happens, we need to deal with each cell-conditional contingency tables as separate data. Secondly,the
category factor explains about 63 percent of the total internal covariation (when cells are not separated), a value larger than the effect of all the formal experimental conditions combined described in Section 4, indicating a fair efficiency of our dyaà categorization.
At this juncture, it may be of some interest to look back at the formal experimental conditions , and see in vlhat proportion each experimental condition contributed to the creation of different categories. The result
is shown ir. Fig.20. Let us first look at the culture part. About Japanese,
CO~ dyads occupy about .68 which is the largest among the five cultures,
while their COOP is also the largest even though in a much smaller scale , leaving the rest categories to occupy only .26. The Anglo-Americans
entirely miss COOP. Belgians are ~racterized by the smallest COMP and
the largest NEUT among the five cultures, while Greeks are of the largest
MART.
Cáncerning the grade effect, it is noted that the proportion of COMP increases with grade, while the proportions of COOP and SERV decrease. The proportion of EXPL remains nearly thesame. Moving from grade to display
Q,II,CI l-x.p...
, we see that the proportion,of COMPv _ _1\inc~eas~ from the Single Display
,
conditionjto the Double Display condition, !ilJ'lo LQQlaitm·
me-
sáme, and theother categories are compressed nearly proportionally.
Now let me proceed to our categoeization of players Version 11 as a preliminary to the model buîlding. The major distinctior. of VersionII from Version I is that the players are now classified also by their
., and then goes done to ] for
as weIl as their CR frequeneies. Another minor alteration is that the elassifieation of players is no longer based on the Post-stage data alone. It is done, instead, based on various parts of the data, depending on the purpose of analysis.
Now note that the maximum possible NR is linearly goes up with ]
CR up to CR
S
2"
Max CR]
2"
Max CR< CR,::; Max CR as shown in Fig. 21, 50 that we may only need to splitaeeording to NR the players with the middle range CR if we want to keep the number of player eategories reasonably smalle The eomparable
elassifieation sehemes for 10-trial bloek and 25-trial blöek are shown
in Fig. 2] with eategory symbols. Note that the + players in this seheme
eorrespond to Ö , =.to! and.!!. and
!:
to ~ in Version-I, if we negleet thetrial bloek upon whieh the elassifieations were made. From now on the
eategory to whieh a player belongs will be referred to~as the state of
the player on the trial bloek where the elassifieation is made.
Fig. 22 shows the relative frequeneies for the four states plotted against 10 trial bloeks.The curves demonstrate that the major state changes
oeeur among the three states --,.!!.-and~, the frequeney of the first steady
'increases andthe other two steadily deeline, while the frèqueney·of +
remains almost the same, Corresponding to the state change for individual players, we ean also plot similar results for dyad eategories represented
by the combination of the states of the two players. Fig. 23 goes along
weIl with Fig.22 as expeeted, though it also display player interacions
as weIl. Fig. 24 shows the state transition probabilities for players
the tendency to stay 1n the same state 1S the highest for all states except
L. In particular the transition- ~ - remains extremely high, suggesting
that the -- state operates as an absorbing state. On the other hand,
however, the increasing tendency of + ~ + and H ~ H imply that a sort of
response fixation takes place with + and H types of response modes as trial
proceeds. Compared to these other states, .the state L àpparently plays a
special rolè such as the middle-way station from + and H to - • This is,
of course, not much surprising, as the state L represents, by definition, the really random choice, and as such, and L-fixation behavior can hardly takes place. Now let us go a little deeper and see how these state transitions are affected by the state of the opponent. Fig.25 shows the results only
for the transition between the first and the second 25-trial blocks, as, in general, there is not much trial effect in these patterns. ASflearly seen, the transitions from the - state are very little affected by the opponent's state. The other transitions are more or less affected by the opponent's state, but the particularly noteworthy is that the - state of the opponent significantly increases the transitions to -, which, coupled with the absorbing characteristic of the state-, apparently precipitates the general rush to competition. This attraction effect of
. .
the opponent's - state, however, gradually goes down with transitions
from + and H as trial proceeds though not shown in Fig. 25, and, in
particular, reaches effectively the zero level for the transitions
from + in the period from the third to the fourth trial bloek. Instead,
however, the attraction of the opponent's L state proportionally increases,
indicatinj that the response fixation for + and H disèussed above is no
Now let us shift our attention slightly, and examine how eaeh player's initial response tendeney, represented by his state in the first
10-trial bloek and ealled'the original state, determines his later states.
The relative frequeneies öf eaeh state in every 25-trial bloek for eaeh
original state are plotted in Fig.26. Sinee the first 25-triál bloek ineluded the first ten trials used in defining the original state, there is no wonder that the same state as the original state oeeurs most frequently in the first bloek. Nevertheless, their values are far below 1.0 (except for the state - ), implying that a drastic response shifts take plaee during the period of 11 - 25 trials as suggested by Fig. 22. As trial proeeeds, most players move to the state - irrespeetive of their original states, even
though a slight tendeney remains sueh that the g~eatest population of eaeh
state is, as a rule, oeeupied by the original inhabitants.
Now we ean also define the dyad original states by eombining the player original states. The top graph in Fig. 27 shows the proportion of their remainig in the same state as the original dyad state for a few samples. The eontinuation of the state -- is the highest over all the
trials, then eomes the state ++, though its proportion nearly steadily goes
eown. The remainig dyad states shown in this graph,i.e., HH,LL and HL
show a preeipitouJdrop in the first transition. Thethree lower graphs in
I
Fig. 27 show whieh dyad states they have mainly moved t~. The superiority
of the state -- now eomes as a no surprise, but the faet is interesting that the dyads whose original state was LL showed the least resistanee.
7. Simple State-Transition Model
model ,obviously, cannot be very much sophisticated. So we decided to try out a simple state-transition model which predicts state transition
probabilities given the opponent's current state. The model involves the
following parameters: 1) The remoteness
Q
from one's current state to hispossible state in the next trial block. The value of remoteness is symmetri-cal between any pair of states, and the remoteness between the same states
is normalized as one. 2) The attractiveness K of each state. 3) Tlie effect
of the opponent's current stat~ , which is supposed to enhance the
attractiveness of the opponent's current state by being multiplied to the
corresponding!. As the states are defined under Version 11, there are
four K's, among which K+ is fixed to one for normalization. Likewise, there are six free D's, those representing the remotenesses between two different states.
For convenience, let me show the equations connecting these parameters in a form óf an example, the principles of which can be easily
extended to general cases. Suppose that one's current state is Hand the
opponent's state ~s
-the form:
Then there are four transition probabilities P of
P(+jH,-)
=--!:L
lra-
+t.+·H-D+H
I . ,-
, ,
P(H/H,-) =--!lL
/LH,- +E
H·H -DHH'
,
(1)~/t
P(LjH,-)
= +( DHL / H,- '- L;H, P(-/H,-)~I<-I[H
-
+E_.
H _
HL ' , ,where the normalization factor
LH _
,
is expressed as(2) + +
and
t's
are error terms such as(3)
LEi;H,-
= 0i
At the present stage of our model building, the purpose of the model
1S to estimate these parameters corresponding to each of the three transitions
betw~en the consecutive 25-trial blocks, whereby the left hand side transition probabilities are estimated from the corresponding relative frequencies.
The parameter estimation was done through iteration by minimizing
the disturbance (or stress) measure,
$' ,
defined as(4)
s
=
L
k, i,j
N' .
1)
where Nij 1S the number of players who actually made the transition from i
to
1..
The result of the parameter estimation is shown in Fig. 28.
Unfortunately, however, the author's stay in Eindhoven has ended when these first set of estimated parameter values were obtained, and their reliábility
(or unreliability due to the risks inherent in the iteration technique)
has not yet been checked. Therefore, the values shown in Fig. 28 and the
conclusions drawn from them are only tentative.
Admitting that, we can observe the following interesting facts:
The attractiveness K and the opponent's effect~ stayed fairly stabIe,
The parameter that underwent great changes we re the three remoteness measures which did not involve the state L, endorsing the hypotheses that the response fixations would be represented by increasing remoteness and that the state L played a special role of a middle way station.
Let me emphasize again the tentativeness of this conclusion and point out that the reason is not at all clear at this stage why remoteness rather than attractiveness changed with trial; whether it is really what the data dictates or an artifact due to the structure of the model.
REFERENCES
McClontock, C. G.
&
Nuttin, J. M. Jr. Development of~
competitive game behavior ~n children across twc cultures.
J. expo soc. Psychol., 5, 203-218, 1969
Toda, M. Covariation analysis : A method of additive data
structure analysis for frequency taóles.
Column Flayer
a b
Ro\\' Player
a b
Fig. 1 The payoff matrix.
~he left-hanè number in each cell is the point for the
row player, and the right hand number for the column player.
Table 1 Sample eize Greek Culture 13elgian Japanese Ss 220
I.
I
108 I II
192 I i I I 168 II
180 ! 9054
9684
110 Number of dy ad anglO-.dmerican rlexican-Ameriean ~ - ~ - - - _ . ~ - _ .I
I
-_.,~.... , 868i
---1
Table 2Standard deviatiolls of the distributlon of the intradyad differenees
---- --- - - 1 i'<"umber of Ss I
---=--t
192 [!
4.25
4.15
IFirst 251
Last 25 , S.D. I S.D. Japanese .n.nglo-··Ameri e anin the eompetitive responees in the first and last 25-trial bloeks.
r
-I
lIJulture
Belgiar.
4.16
4.,;4
îGS
III
1.0
IV III p 00.9
Pi III OJ S-t0.8
Cl> > .r! ~-" .j...l0.7
.r!.
.'
-.j...l/ '
.-Cl> J)." .. -Pi ~..
'.
-Ei0.6
" . •
-c:a
---0.
..
-() / , - --'H0.5
~-0 Cl> .j...l0.4
cu Cl:<T
•
Japanese X Greek À Anglo-rl-merican 0 Belgiana
Chicano 2 46
GraàeFig. 2 Culture X Grade effect.
( Display conditions pooled )
5
4
3 2 1 Belgian ....
....
.....
...
....-./~~/
~
54
3
2 American 1 54
3
2 Greek 1 5 43
2 Japanese.
...
--.
-
....
...
IJ',
I I/
I 11.0
<ll III I:l0.9
0 Pi III Cl> S-t0.8
Cl> > 'r!0.7
.j...l .r! .j...l Cl> Pi0.6
.S 0 () 'H0.5
0 Cl> .j...l0.4
m p:; 20 - Trial Bloeks{Q ~ 0 'n +> .r! 'd ~ 0 UI 0 111 (lJ :>:. ~ cU (lJ r-i ::- P< .r! 111 +> .r! 'r! 'Cl +> (lJ 0 P< E: El +> 0 0 ~ (lJ ~ (lJ 'r; +>
a:
(lJ (lJ 0 .0 ~ (lJ H (lJ 'H ~.~ .r! Q 0.20 0.10 O. 2 4 6 2 4 6 2 4 6 24
6 Graàe Anglo-American Greek JapaneseFig.4 Display effect for Culture X Grade.
Oràinate represents àifferehecs {n the rates of competitive responses between the Double Display conàition and the Single Display conàition.
·25
Greek -~ N NVJ v..' +-+-o \.11 0 \.11 0 \.11 1 I I I ' 1 \,
\ \ Chicano.
... "",--,"
.,....", Y . I,
,.
, " , ,".....","'''' I • PPD o -' -' 1 \.11 0 \.11 +- 1 I I "... 1 1 1 -'-'\.11 \.110 1 1 NN \.,'1 0 +-\.11 I I 1 1 1 ,,
,
.;-_./ "'- ,. .. " \ '..,.e'--\---+--'~--1 1 V<VJ \.110 VJ \.11 I VJ +-\.11 \.11 I 1 N \.11 1 - ' \.11 I PPD ,'\,
\,
\,
\,
,
\ PPD o , \.11 +-,-L~~~~~:
iL\.11 \.11 \.11 4-- 1 1 1 I. - ' o,
- ' o,
,
N C,
N o 1 VJ o Anglo-American,
\.,,, o ''''~-. .,.; < .~, ... +-\.11 1 +: \.11 1 VJ \.11 1 N \.11 1 \ \,
,
,
\
~....,.....
~-:x'" N VJ \.11 \.11 1 ,,
,
\"'
,,
\ \,
\
,,
'
...PPD
- ' \.11 1 - ' \.11,
Belgian . __.,\-.-..- ..--- -tPPD
" \.11 +-1 \.11 +-1 I - ' o,
- ' o,
1 N o,
,
N o,
I ,.,
"---,~---_.~.~,-~, ._~_.
--.\---I 1 I +- VJ +.- 0 0 - ' I,,
~
,-
' ,1,
- - - --'--+--4'!-=-"~""~--r-,,~m . .~--~~~-,---+~ • • 1.15
.10 .20.05
.00 .0) IJapanese ~g
.25
(l) ;::s ')0 a' • L-U> H 'H ~.15
rl +> (û .10 rl <l> ::r;Single Display condition -- - --Double Display condition Fig. 5 The probability distributions of PPD under the two display condition
o
5
Hean
10
15
Pooled matrix all SI Rl N Decomposed matrices
,
s' all . 1 r' n' 1 a' , s' 11 1 r' , n' 1 s' r' I 1 n' s" rIf = a" - 1· 1 11 -n"WT Total internal covariation
Internal covariations . tt'
w:
External èovariation*
W = W -T W.~RESIDUA L
---TOTAL
I NCOV
=
1444.6~EX C0 V ( T
x
D XG
'i.C )
=
Z
5 9.7 8EX
C
0 V ( DYAD )(
~)
=
~ ~2 8.65Fig.8 The overall results of covariation analysis.
DISPLAY
EFFECT'
2'C=
7 5 9.7 8(E XC 0 V )
'rrial block',
. . ',...
•....
Grade 2,4, 6
Culture Japanese ~nglo-Arnerican ~elgian Qreek• •
•
•
CU l TURE
EF
FE CT
27t
=
759.78 (EXCO V )
Fig.j10
Covariation analysis Culture effectDisplay 1t II
TRIAL
EFFECT
2x=759.78
(EXCOV)
I/
GRADE
EFFECT
1C
=
7 5 9. 78(EXCOV )
Fig. 1~ Covariation analysis
1.0 0.5
-
---
_.---
....
--
-..---..
"_
... • ' •• - - - - ;-S- ... A -~-~ ..&~~~~~-:::-:--_-"':A>":::::':-:-'-::-_~_~_·_-:"'_~'::'A~-=-_-_- ___ 0 - - - : ..~::.._4" - · A 6 - , -..- " 4 ,,' ~-.., --~ af ter eell 1 eell 2 eell-3 eell 4 ~ CIl ç: o '.-1 .(..I '.-1l
]
o.
0L_-.L._--,__
~----,--~_-.L._----4---~--2 3 4 5 6 7 8 9 10 10-Trial BloeksFig. 13 Conditional CR probabilities given the preeedingeell
plotted against trial bloeks (Runlength-pool)
1.0 0.5 0.0 .' ~'----~
-
--
:.- --~ 0- -_- _ -.:10- - -A-:" .... ,.-2 Runlength_--0
4 - - - 4 af ter eell I eell 2 eell 3 eell 4Fig. 14 Conditional CR probabilities given preeeding eell
1.0 3+
"
C eell LC ,,
af ter..
eell 2"
""~ 0.5 À eell 3•
eell 4 0.5MC
1.0Fig. 15 Conditional CR probability given the preeeding Cell and
the length of Run for the two players.~ Trial-pool)
.30 .25 .20 • 15 • 10 •05 -\ \ \ ~
.
o
CÖÖP NEUT COMP I I I I+:'-w +:'-OU1 - I I I I WN OU1 I I I I N -OU1 I I I I 0 -lJ1 I \ J 1 -O ' · + : ' - I -O U 1 I I I NNW 0 U 1 0 I I I W +:'-U10U1 I I I PPD/ l
/
co~ (20~
I
/f----
---J
/ I / I /NEUTI
EXPL (96 ) / / (72)I
~11'>/
/ ///
r-/ -
f' /
./
./io/ ; , r , " I.,//
i
/ .. -'
I,:
/
: ' / / j :
/ L , / ' S E R V !L
/
----.~.--_.~,j---L
caap
(
J4 )!I
,
(
42)I
: MART ( 8 ) iI
75l
CR of LC X 46 45•
l
N 30 29a
o
a
---a
29 30 ~ 45 46 N CR of HC--
x
75Fig. 16. Categorization with dyads.
The numbers are of dyads belonging to each category.
All the Mart dyads locate in the upper left quarter(the
coop
( 0 x 0 ) 1 . 0 0 1 - - - ~---___r first 25 trialso
last 25 trials 1.0 ( Nx N ) 0.0 1.0 ~ "f
'..,l'n
- - - - -
.
1.0 0.5 1.0 0 SERV (o
x N ) last"
, ,
,
N NEUT firstI
'--:---::-_0::---_ _1 0.0 0.5 1.0~'Y
'~, '"4 0.0l,
,~
~
0.0 0.5 1.0o.
00 ~---4- ~ 0.0 N NRate of competitive responses (more competitive player)
Fig. 17. Conditional CR probabilities g~ven the
.0
af ter eellpreeeding eell for eaeh eategory. Estimated from
A
af ter eell 21.0
l
I
I
I
l
I
I
1 L - ' . . . ---J. 0.0 0.5 1.0 MART (0 x X ) 1.0J
x
I
0.0L,_-+-,_
0.0 0.5I
I
0.5 I i I,
o
1.0 ( N x X )I
t!
1.0 EXPLi---l
0'5~
\1
I
IV
Ir
Ii\
I
I
0.0L--~~---J
L
__,--0.0 0.5 1.0 0.0 N al po .~ +> .~ +> al Pi -f.:l o u al po .r-! oP .~ +> al p., El o u X 1.0 COMP (X x X ) XI
I
i---Yl
i .
~~
I
.
. I .
\
11 0.5L
~,.
ih
I
I
.1I
,
I
i
0.0L----.
J
L __
'_..L-.-_ _J
0.0 0.5 1.0 0.0 0.5 1.0 XAfter-Cel1 til al '..-1 .~
·f
-r4 '..-1 ..0 til ..0 o I-l p., p:: ü After-Cel1 2 til al 'j ·r--..-I r4 • ..-1 ..0 cu ..0 o I-l p., p:: ü ---_._---_.- ,-/.0 CR ProbabilitiesSERV NEUT COOP MART EXPL COMP o
"
o
····f
/,,0i
1 ~3 'f . ~ I After-Cell 3 After-Cell 4 /. 0 , -Lcl·
i
ILC
i i• S "
• S-CR Probabil~t'.L ~esMe.
Japanese
MI \'---"'-"-~'-, -.----.~.- -- - ...•-..-.-.-- ---.- - " , _._-~
"'·'1
NEJJA! I
COOPi SERV! UT
~I
EXPLI
COMPI I
:..L-.-._._ _
--l_~
..__.. .._._ _'.' __'<~
' " •• _~
• •" , , _ _ _. ._~.,,>c _~_
.._._, __
_ _' _ __.. .. /
\ I 1
\---r···..--··--..----·---..··-..··---··--·--···-·.._..-_.---..
~_..
~~~~can ~UT
I
!
ExP~
I_.
~~~
__.
._
..__
, ..
Belgian
Greek
I
I
-.__._-._.-..__.
---r--·>c..
'"···,,·---~-.,·~~··,··~"'_·-..,-.-.-..
--.--~.."-"---._,,
~-._-.,
····1~
SERVI
NEUT!
EXP:~o~-
...- - , - - - - -..----.l
~J ~ERJ~;
_ __rI:;~==
__
=-=~I-:·~=_~;_::·-:=:·===-
__
~~-_-=._=:::1
~
:----·r---..·---..
-·-T~T....·-.."·-·-..
_··~·_····~··,--·..._.,...,._"...
"""0"'''''' ...'>."...,'I' .'...." ....0,
I
I
R.j
Chicano
'---'-
Pi SERV----!..____
1 NEUT 'T EXPL COMPl .____ ___ _...J.. ••_ _._ ~•• . _._ __ _•. .
Grade
Second
,.--._-,....,.._-_.. -- ...----..-... -- -_. -.'r'-~- '~" -- ...-. -. -.- .. -,. .. ", r '
I
I ,IMi
\ coop
l
SERVI
NEUT;~j
EXPL,_ _._.~ . ._ _...L _...._ _. ._ .• .•,. ...•. .1 __• ... ... • ..._ CaMP
---_.-_
....--.. Fourth SIXTH Display Single DoubleiI
1 - - -·--!Ml·"·----"--· .. ··-,,·
_._. ·..
···r..·"···_..··"..
~··'"···_·_···-..
_"._~_-_._
_"..-.- "
I1" '1;IRI
lL§.F;By~~~2:_
.._._ __..
_--.l:.l ..
".~~~~
"'__ _._
• caMP~._.·._.._ _ ..v"'" ..._<. ....c] .
'r _."..'"."-~.-.~" ,. _._ _ -._ _.--" I " .--..--- "'-..----.- -._-. - -., __.._ ..-- _~---~..- - ..-- _ - ._ - -..;,l'\
NE~;
I 1~IR UT RI E X P L ! CaMP . lfiv'Tl
i
_ _ _ _..._ _"._ _"'...""',...~...,.."..-""'" • __" ....-..."._'...,.' .. ' -...~....,.~".~,~,...,.__~..."~··'v'~..-""'_""'~'..~,'"·,,~,',--"-,, "'...",..- ..'" •....··'··;"T\h"" "~••·,'., ...,...,'t·.,,...,·..._,,-.·,·...-..."~,.".,,,..,.,,,,_.,,,;..r.'~~-"" N'''<'-·-1---·"·..
-r'···-··
w._ .. -. ----->
T-"--'-'-"-~~--"'--- .. ",." - - - -- -- - - . - - - . - . - - - .-. r I1~11
iI
I I R . 'L~~~L..::RV~
..J __.__ ...
N~UT
I
Tl. __
EX~~
. . .» ..L.
CaMPQ)
·03
:>...
4J C'il r-I·2
. Q) p:::. I -
_
....--.>---.;c.---_ _
---.I>---<---
----0
G+
o
1.r
IQ-trial blocks6 7 , .
10Fig. 22 Relative frequencies of personal states plotted against IQ-trial blocks.
IQ-trial blocks
3
5
7
V
[J/ i'
2t; )(
I
/-t-I
/
~'--"H
- ;---_.._~\
L
cR
\ \ {D ) l7
o "-
J '1 '"\. 10Fig. 21 Definition of personal states for 25-trial bloek and lO-trial bloek
.I
s
0 -'T--t 0 ,* -A ~r L1--}L
0---. 0-.. ---0L
• __. _ • ~_.. J 1-71. -) 3 -""/r~H
H -) H ,}S-i._1
/.0 .-·1
.?,·1
i- -» + til'b
----.--.
Q) .,. • .-1 -IJ • .-1·s
~. ... .. -.-I .n Cd .n .~ .. 0 l-i p.. ~3
. -t -~ !... 0 -.-I A -IJ .4 , • .-1 ...,---.~--..Ll#'/ til . .2. ~ Cd l-i 0., E-l '>._.4_.
I .4- "'0 •--,4~-o
"
.J.!(
0 ) I. () •-
5
I Il. H L\ Lo
A,g
1/'f!
I
1
" '/1!
/,I ,./l
,.1)1 0--- .? .... IJ'---~,.~-=(~
__=~;-''-:''7".~_~<.-~---'_ .- - _ .+
H
L
L
L
H
+
t
,/
~--1---41
i It--=...:...:":::: ,/._./
0 - - ..-l!.:/~,
P
.
,...
, / . / I .... , , " " ' , .. I ... " " '~ , ....-11---/0
'
0- - - ..' ,'. ,,r-.,.~ , ' /~--
....
' i ~ 'I / '/ / ' / ____.0..._ ..__··_ ...+
H
L.
H
'\\
P
4 \ ,A,\\\
',\ IK
/
....\ / ~, ' \ j. --&..." , \' ,," 1-"~=A~_
-1~--'H
L
''--.. '" / ' . / '7--->
-t
s,
CIl Q) .~ -IJ .~ .-I .~ ,.Cl Cil ,.Cl o '"'"P-S .
~.
.~ -IJ '''; CIlg
'"'" Eo-<S..2 --"
Fig. 25 Transition probabilities of personal states for the transition from the first to the seeond 25-trial bloek.
P ( S2 / SI' Si) Si : personal state in the i-the trial bloek. si : opponent's state in the trial bloek 1.
[)
I
I
'I ,.
~<:.,.
.,/,/
LH
• .J,. • .. ->~--~.:- ~i .~ ~ ~,
~~ o Trial bloek 4+
.. ,I.
/
o,
Trial bloek 3;'/6
'/.i~
JI",i, ... !{/// --... .1.--- .._LtffH'1
..
,'" ...-. ~L.. ..._ - - " ' _. . o -.._·_·..~/ , ....~.:-'0 . - .. __0+
H"'"L.
i 1 i , .9
I/
Trial bloek 2L
:-( -I-Trial 'bloek 1?
r
/
I
"
/
I
I t>-I \I
/
\ / \ \ ,4,. It
(,/
I , /'0
I
/'
.~
\ I \ / ' ,A-. i' ,_.:;;. \'' \ //' \y~~
~
I
' -~
.. " - , -"..-
':"Ié/~'
I " ' / \.
~. I ''o, / ; /~(
• "';._A;:.:/ I I .L>' / , - - I AA' _ _ - " _,0 .. 'v' 0 '"I.~
_ _-0 ._'." "L..---t}---o.<::----=.---...__....:...
L
+.,'...
H
L
1,0 Q) ':;' .~ -IJ (Ij .-I Q) p::: CIl Q) .~ t) ~ ~ .!; I-0" Q) '"'" 4-l Current stateFig. 26 Conditional relative frequeneies of states given original states for eaeh trial bloek. 0
+
r or;j
t
-I.Ii .o.S
r
I ior;
cin:,L [(Jrh";>::: d•
+
+
--~ f- T 0H
H
-';> H H ~ L L - - } L L.-X ~> t:I HL.
---'.> HL
Ati
- ' ) ~L 0L.L
~ )<1.L
-4HL
0L.L
- 7L,-•
/-IL.
- ) 11L
0HL
-7 AHL.
- ?L-1\
HL
-_.)-L
I'i
I D.Sl
I
/.0 IO-trial bloeks IO-trial bloeks o HH-)•
Hf1-è}
A H f-l ---}x.
Hf1--}Hf-!
LL
f
I , <~ j ,-,x"
-~~
~X
---)(K·
..,.
I>K
H Llk
J.sx
o
~__.. ._ _J ~_.~~__ J-7L/. ~---+', -l-tolt
1.--->330
I-l
,I
I6
t-I i I QJS
1-I ;j \ ..-l cd I ~ f~ ~I
QJ3
~
.IJ QJ ~ I ~i
(Ij 1. ~ • .0. P-t I I 0 ~ 0 i , ! I ~ i ( i II'
,/
/
10 .~ ;. / / 0Dli
-..//
iJ ,////Fig. 28 Parameter values estimated through
iteration in the simple state-transition mödel