TABLE 2.6 Construct Validity of Sociometric Status Groups: Summary of Findings
TIME 2 Same Sex Opp Se
.50 .09 sd .24 .15 .24 .14 G irls m ean .47 .06 .47 .13 sd .26 .10 .25 .15 t .93 2.16 1.18 -2.04 P .35 .03 .24 .04 Boys m ean .16 .58 .21 .60 sd .19 .26 .22 .24 G irls m ean .16 .61 .20 .55 sd .23 .25 .22 .28 t .15 -.81 .21 1.56 P .87 .42 .83 .12 Boys m ean .34 .32 .29 .31 sd .18 .20 .15 .19 G irls m ean .37 .33 .33 .32 sd .19 .21 .19 .22 t 1.44 -.34 -1.95 -.58 P .15 .74 .05 .57
TABLE 3 .8 P e r c e n ta g e o f C h o ic e s R e c e iv e d in e a c h LITOP c a te g o r y fro m S am e a n d O p p o site se x p eers, b lo ck ed by sex.
A lthough som e exceptions are m ade by som e children, th ere is stro n g evidence th a t children of this age adopt the attitu d e th a t they do n o t like to play w ith' o p p o site sex p eers in gen eral a n d apply th is in a n o n - discrim inative m a n n e r to th e ir opposite sex classm ates. On occasion 1 15.7% of resp o n d en ts gave a negative rating to all opposite sex peers, while no resp o n d en ts gave a blanket rating (of any sort) to sam e sex peers. On occasion 2 17.9% of respondents gave a negative rating to all opposite sex peers, while only .07% gave a blanket (positive) ratin g to sam e sex peers. FCP sociom etric classification system
Table 3.9 illu stra te s th e way in w hich assig n m en t of su b jects to th e FCP sociometric classification m ethod categories is affected by th e u se of sam e
or both sex forced choice ratings. Both Sex P Ç OCCASION 1 A N R OCCASION 2 C A N R Sam e P 7 2 64 2 3 17 2 56 0 3 Sex C 0 0 0 0 0 0 0 3 0 1 A 1 0 109 6 26 1 1 106 5 39 N 0 0 14 4 3 0 0 2 1 3 R 0 0 2 0 9 0 0 1 0 13 T a b le 3 .9 A s s ig n m e n t o f s u b je c ts to t h e FC P s o c io m e tr ic c la s s if ic a tio n m e th o d c a te g o rie s u sin g sa m e & b o th sex d a ta
Only h alf the subjects are allocated to th e sam e category, of the 5 available, u sin g b o th sex an d sam e sex ratings. Many of th e subjects classified a s po p u lar u sin g sam e sex ratin g s are classified as Average u sin g b o th sex ratings. Many of th e subjects classified as Average using sam e sex ratings are classified a s rejected u sing both sex ratings due to the addition of opposite sex ratin g s w hich are less positive in acceptance an d m ore negative in rejection. To assig n su b jects to sociometric s ta tu s groups u sing b o th sex ratin g s w ould be to identify a s rejected children less th a n a q u a rte r of whom are rejected by sam e sex peers an d m ore th a n h alf of w hom are averagely well accepted by them.
3 .3 .5 S tab ility o f so c io m e tric s ta tu s group classificatio n
The stab ility of classification obtained w ith each of th e m eth o d s w as m easured by com puting th e percentage agreem ent betw een th e two tim es of a sse ss m e n t a n d th e n correcting for ch an ce agreem ent w ith C ohen’s Kappa (Fleiss, 1981). The stability of each m ethod is fisted in Table 3.10. Percentage agreem ents range from 44.49% (Asher & Dodge, 1986 m ethod) to 68.1% (FCP method). The range in kappa values is sm all (.30 to .43). The v alu es recorded in th e z colum n in d icate th a t all th ese classificatio n m ethods produce a n overall value of kappa w hich is statistically significant. However applying th e criteria recom m ended by Fleiss (1981) only th e following m ethods have k ap p a values w hich can be regarded a s fair a s opposed to poor: Peery (1979), Newcombe & Bukowski’s (1983) application of Coie et. a l.’s (1982) m ethod, F rench an d W aas (1985) a n d th e FCP m ethod. While differences in test-retest reliability values betw een m ethods are m arginal, b o th th e h ig h est percentage ag reem en t an d th e h ig h e st overall kap p a value w as obtained for the FCP method.
Percentage Kappa
METHOD A greem ent
Peery (1979) 5 3 .9 5 .4 0 12.2
Newcombe & Bukowski’s 5 1 .5 8 .38 11.6
(1983) application of Peery’s (1979) method
Coie et. al. (1982) 4 9 .6 .31 9 .2 9
Newcombe & Bukowski’s 5 6 .6 7 .41 12.63
(1983) application of Coie et. al.’s (1982) method
Newcombe & Bukowski 6 6 .5 6 .3 4 8 .8
(1983)
Cole & Dodge 5 7 .4 4 .33 8 .8 8
(1983)
Newcombe & Bukowski’s 5 7 .0 9 .3 9 11.1
(1984) application of Coie & Dodge’s (1983) method
Newcombe & Bukowski 5 7 .0 8 .36 9 .9 3
(1984)
French & Waas 6 1 .5 9 .40 10.2
(1985)
Asher & Dodge 4 4 .4 9 .3 0 9 .6
(1986)
Sabom ie et. al. (1990) 5 2 .7 5 .33 8 .6 8 Morrison (1981) 4 8 .4 3 .3 0 8 .1 5 FCP Method 6 8 .1 .43 9 .2 4
T able 3 .1 0 T e st - r e te s t relia b ility o f so c io m e tric s ta tu s g ro u p c la ssific a tio n s
3 .3 .6 V alidation o f S ocio m etric S ta tu s G roup C lassification
O n Occasion 1 and Occasion 2 separately, th e peer a sse ssm e n t profiles of sociom etric s ta tu s g ro u p s created by each classificatio n m eth o d w ere examined. D ata collected on each occasion w as analysed separately by way of replication. For each m ethod scores from th e 6 item s of th e G uess Who, transform ed as described below, were analysed by a one way MANOVA with planned co n tra sts betw een each of th e extrem e sociom etric s ta tu s groups identified by th e m ethod an d th e Average group. W hen MANOVAs yielded sig n ifican t m ain effects, c o n tra sts betw een th e g ro u p s w ere exam ined se p ara tely for each d e p en d en t m easu re (the 6 G u ess Who item s). For sociometric classification m ethods using sam e sex ratings, sam e sex G uess Who scores were used. For sociometric classification m ethods u sing ratin g s from th e whole class. G uess Who scores were compiled from th e whole class.
in te ra c tio n s w ith e ith e r gen d er (F^2 468)"^ '^ ^ ' p=.38) or y e a r group (F(24932j= 1.02, p=.43). Because of th is and the reduction of cell size w hen ad d itio n al in d ep e n d en t variables were included, th e MANOVAs reported here include a single independent variable: sociometric sta tu s group.
E^valiiation o f A ssu m p tio n s
R esults of th e evaluation of assum ptions led to the transform ation of Guess W ho v a ria b le s to re d u c e sk e w e d n e ss a n d o u tlie rs . S q u a re ro o t transform ations were u sed on th e sam e sex scores collected on. O ccasions 1 an d 2 on all item s except 'cooperates', w hich w as n o t tran sfo rm e d on O ccasion 2 a n d on w hich a Logarithm ic tra n sfo rm a tio n w as u s e d on O ccasion 1. S q u are root tran sfo rm atio n s were u se d on th e w hole class G uess Who scores collected on both Occasions 1 an d 2.
G roup sizes differed considerably acro ss th e sociom etric s ta tu s gro u p s identified by each classification method. U nequal n w as dealt w ith u sin g the regression ap p ro ach in SPSSX MANOVA. This is th e m ost conservative of th e available approaches and avoids giving heavier weighting to larger cells. In th is stu d y th e largest group on m ost m ethods ten d ed to be th e Average g roup w h ereas th e extrem e sociom etric gro u p s, w hich were of g rea ter in te re st ten d ed to be sm aller. In a n u m b er of in sta n c e s th e req u irem en t th a t there should be more cases th a n dependent variables in every cell was not met. To deal w ith th is problem any group w hose size was less th a n 10 w as deleted from the analysis and a ‘D’ entered in Table 3.11. In testing for th e a ssu m p tio n of variance-covariance m atrices T abachnick & Fidell (1989) advise th a t if sam ple sizes are un eq u al an d Box’s M te s t is significant a t p<.001 ro b u stn e ss is not guaranteed. The significance of Box’s M reached th is level in only one an aly sis (th at of Newcombe & B ukow ski’s 1983 ap p licatio n of Perry’s 1979 m ethod on occasion 1); o th e n ^ s e it ranged from .002 to .627 so it ap p eared re a so n a b le to a ssu m e ro b u s tn e s s , especially a s Pillai’s criterion was used to evaluate m ultivariate significance. Table 3.11 show s th e F values and significance levels for th e m ultivariate te s ts . All c la ssifica tio n m eth o d s p ro d u ced so cio m etric g ro u p s w hich differed significantly in th eir G uess Who scores o n b o th O ccasion 1 and Occasion 2. The resu lts of the planned co n trasts analyses illustrate th a t for all classification m ethods w hich identified a n Average (or in th e case of Morrison, Flat) s ta tu s group, significant differences on th e G uess Who item s were obtained on bo th Occasion 1 an d Occasion 2 w hen th e Average group w as c o n trasted th e p o p u lar group a n d w ith th e rejected g ro u p . Hence all m ethods in Table 3.13 down to Peeiy (1979) consistently identified p o pular an d rejected groups w hich significantly differed from th e Average group on the G uess Who m easure.
M ETHOD O ccasion Main Effect Planned contrasts (F ratio) (F ratio) PvA CvA NvA RvA
Coie et. al. (1982) 1 4.39 *** 7.16*** 1.20ns 1.32ns 5.53***
2 4.52*** 3.60** D 1.20ns 11.38***
Newcombe & Bukowski’s 1 4.48*** 4.51*** 0.58ns 1.02ns 6.27***
(1983) application of Cole 2 4.89*** 7.26*** 1.07ns 0.48ns 5.77***
et. al.’s (1982) method
Newcombe & Bukowski 1 3.90*** 3.86*** 0.40ns 0.90ns 11.12***
(1983) 2 5.91*** 7.55*** 2.47* 1.94ns 14.32***
Coie & Dodge 1 5.19*** 7.28*** 1.24ns 1.33ns 3.62***
(1983) 2 4.77*** 6.22*** 0.75ns 0.80ns 13.34***
Newcombe & Bukowski's 1 5.27*** 7.35*** 1.36ns 0.83ns 11.22***
(1984) application of Coie 2 6.21*** 10.26*** 1.47ns 1.54ns 15.34***
& Dodge’s (1983) method
Newcombe & Bukowski 1 5.20*** 3.89*** D 1.45ns 11.35***
(1984) 2 5.72*** 6.40*** D 0.97ns 10.70***
French & W aas 1 6.78*** 7.57*** D 3.12** 11.09***
(1985) 2 8.52*** 14.42*** D 1.64ns 12.27***
A sher & Dodge 1 4.32*** 6.93*** 1.24ns 1.33ns 3.62**
(1986) 2 3.42*** 3.22*** 0.67ns 0 .14ns 7.67*** FCP M ethod 1 4.40*** 4.00*** 0.88ns 5.80*** 2 7.83*** 10.34*** D D 5.23*** (AcvF) (Ac/RvF) (TvF) (RvF) M orrison (1981) 1 3.70** 3.16** D 1.58ns 3.42*** 2 4.11*** 3.08** - 2.31* 5.22*** Peeiy (1979) (1983 (application of Peery's (1979) method
Sabom ie et. al. (1990)
(F ratio) (PvAm) (Pvl) (IvAm) (RvAm) (Rvl)
1 6.41*** 1.8ns 8.44*** 16.62*** 21.84*** 1.29ns 2 5.19*** 2.47* 11.58*** 6.71*** 13.68*** 3.32** 1 5.79*** 1.00ns 9.58*** 9.53*** 21.21*** 1.22ns 2 5.47*** 1.75ns 9.89*** 5.84*** 14.61*** 1.86ns 1 4.13*** 1.24ns 9.30*** 5.72*** 8.76*** 1.24ns 2 4.77*** 1.60ns 9.75*** 6.52*** 12.78*** 1.66ns
*p<.05. **p<.Ql. ***p<.CX)l - indicates no subjects identified in this group D indicates group deleted from analysis due to group size <10
T able 3 .1 1 M u ltiv a ria te A n aly sis o f G u e ss W ho ite m s b y S ta tu s G ro u p fo r e a c h S o c io m e tric S ta tu s C lassification M eth o d
None of th e sociom etric classification m ethods investigated in th is stu d y identified C ontroversial or Neglected g ro u p s w hich co n sisten tly differed significantly differed from the Average group on th e G uess Who m easu re. Indeed the Controversial group differed from th e Average group only o n th e Coie & Dodge (1983) m ethod on Occasion 1 and only on th e Newcombe an d Bukowski (1983) m ethod on Occasion 2. The Neglected group differed from
the Average group only on the French & W aas (1985) m ethod on Occasion 1, while on th e M orrison (1981) m ethod on O ccasion 2 the Tolerated group differed significantly from th e Flat profile group. This stu d y did n o t therefore succeed in validating the Controversial or Neglected sociom etric statu s groups against the Guess Who m easure.
The final 3 m ethods in Table 3.11 classify su b jects into five sociom etric groups - P opular, Am iable, Isolated R ejected a n d U nclassified. T he unclassified group w as composed of subjects whose scores on e ith e r Social Im pact or Social Preference fell on one of th e axes depicted in ch ap ter 2, Figure 2.3. These m ethods do not identify a central group (labelled Average by m ost o th er m ethods, an d ch aracterised on th e one h a n d by a low difference betw een th e n u m b er of positive an d negative choices received and on th e other by a middle, neither particularly high or low, position w ith regard to th e total n u m b er of positive or negative choices received). R ather these three m ethods take all children who receive a lower th a n Average (for their class) total n u m b er of positive or negative choices and classify them a s Amiable or Isolated depending on w h eth er th e difference betw een th e n u m b er of positive a n d negative choices received (social preference) is positive or negative. C o n sisten t differentiation of th e Amiable from th e Isolated group on the combined G uess Who analysis su p p o rts the validity of this distinction, based on social preference.
However th e differentiation betw een Rejected an d Isolated groups on th e one h a n d an d P opular an d Amiable groups on th e o th er receives little su p p o rt from th ese resu lts. While th e Rejected an d Isolated gro u p s bo th differed significantly from the Amiable group on all th ree m ethods on each occasion, only on the Peery m ethod on occasion 2 did th e Amiable group differ from th e Popular group. Using the Isolated group a s th e com parison group produced th e reverse effect in th a t th is group w as found to differ from th e P o p u lar a n d Amiable g ro u p s on all th re e m eth o d s o n b o th occasions, w hereas only on the Peery m ethod on occasion 1 did the Isolated group differ from the Rejected group. A parsim onious description of th ese findings m ight be to say th a t the Amiable subjects can be regarded a s mildly popular while Isolated su b jects can be regarded a s mildly rejected. The reputational profiles of children who are accepted by th eir peers differ from the rep u tatio n al profiles of those rejected by th eir peers, although th ere is little difference betw een children whose rejection (or whose acceptance) is m oderate ra th e r th a n mild.
The signific an t m ultivariate contrasts between th e Popular and Average an d the Average an d Rejected groups were next exam ined for each dep en d en t
m easure. B ecause of significant w ithin cell correlations among the 6 G uess Who item s, assessed using B artlett’s te st of sphericality, stepdown analysis w as used. Homogeneity of regression, a sse ssed th ro u g h SPSSX MANOVA w ith each G uess who item serving as a dependent variable on one step and becom ing a covariate on the next, w as achieved for all com ponents of the stepdow n analyses. The 6 Guess Who item s were assessed in the following order in each stepdow n analysis. D isrupts' w as followed by sta rts fights' as two negative peer a sse ssm e n t item s, asso ciated in previous stu d ie s w ith rejection. D isrupts' was prioritised over 'sta rts fights' because of its greater th eo retical im portance in validating th e rejected group. The rea so n s for th is are explained m ore fully in sectio n 3.1.2. Next 'co o p erates' w as followed by leader' a s two positive peer a sse ssm e n t item s, asso ciated in previous stu d ies w ith popularity. 'Cooperates' w as prioritised over leader' becau se of its g reater theoretical im portance, detailed in section 3.1.2, in validating th e p o p u lar group. Of th e rem aining two item s w hich have