The reliability and validity of the Parenting Scale for Australian mothers of preschool-aged children

Abstract

The aims of this paper were to examine the psychometric properties of the Parenting Scale in an Australian sample and to provide normative data on this scale for Australian mothers of preschool-aged children. The total sample included 1,656 mothers of children attending randomly selected preschools and mothers attending parenting groups. Principal components analysis of the Parenting Scale items supported the original three-factor structure described by the scale's authors. Consistent with previous research, the Laxness and Overreactivity subscales demonstrated high levels of reliability and validity, whereas the Verbosity subscale did not. Specifically, the Overreactivity and Laxness subscales demonstrated high internal consistency, criterion validity and correlated significantly with measures of child behaviour. Whereas, scores on the Verbosity subscale demonstrated low internal consistency, did not differentiate between the two samples in the study, and had small, nonsignificant correlations with child behaviour scores.

Although it is not the only influence on children's behaviour, parenting plays a major role in influencing the development and maintenance of aggressive behaviours in childhood. Ineffective parenting is thought to influence young children's socialisation in a number of ways; for example through modelling, conditioning, or by influencing the security of children's attachment to their parents (Hinshaw, 2002; Rubin, Stewart, & Chen, 1995). The importance of parenting behaviour as a risk factor in the development and maintenance of children's behaviour problems, and the potential of parenting behaviour as a target of intervention programs, has led to a focus on the measurement of parental discipline techniques (Locke & Prinz, 2002; Reitman et al., 2001). Over the past 20 years, numerous self-report questionnaires have been developed to measure parental discipline techniques such as punitiveness, overreactivity and permissiveness. Locke and Prinz (2002) identified 39 such measures for parents of preschool-age children alone.

One frequently used questionnaire to assess parental discipline techniques is the 30-item self-report Parenting Scale (PS), developed in the US (Arnold, O'Leary, Wolff, & Acker, 1993). This measure has been used for theoretical research and program evaluation in the US, Australia, New Zealand and the UK (Bor, Sanders, & Markie-Dadds, 2002; Cann, Rogers, & Matthews, 2003; Cann, Rogers, & Worley, 2003; Gardner & Burton, 2003; Harvey Arnold & O'Leary, 1995, 1997; Keown & Woodward, 2002; O'Leary, Slep, & Reid, 1999; Sanders, Markie-Dadds, Tully, & Bor, 2000).

The PS was developed from a review of the parenting literature and from transcripts of interviews with parents. This scale was designed to broadly and directly assess the domain of parental discipline, and to be easily and inexpensively administered (Arnold et al., 1993). Each item on the scale contains two anchor statements relating to parenting behaviour. Arnold et al. (1993) designated one of the statements a “mistake”, and the other behaviour the “effective” parental response on the basis of prior empirical research, and correlations of the items with children's misbehaviour. Parents indicate on a 7-point scale between the two anchor statements the point which best describes their usual responses to their children's misbehaviour. A strength of the questionnaire is that parents' responses are independent of the frequency of their children's misbehaviour because they indicate which techniques they are more likely to use, rather than how frequently they use them. When frequency-dependent items are used (e.g., with response formats such as always, often, sometimes or never) parents' responses about their own behaviour may depend on the rate of their child's misbehaviour, and this may lead to spuriously high correlations between the two. For example, a parent whose child misbehaves frequently may report using ineffective parenting techniques more often (even though they may also use effective techniques more often) than a parent whose child misbehaves infrequently, simply because they are involved in more discipline interactions with their children.

Arnold et al. (1993) analysed the factor structure of the scale using a mixed clinic and nonclinic sample of 168 mothers with children aged 18 months to 4 years. Principal components analysis of the 30 items of the PS revealed three factors: Laxness (11 items), Overreactivity (10 items), and Verbosity (7 items) (Arnold et al., 1993). Laxness refers to permissive and/or inconsistent limit setting, in which parents frequently back down from their requests for child compliance or give in to children's inappropriate demands or coercive behaviour. Overreactivity reflects parenting characterised by harsh and coercive discipline such as anger, frustration, insults and name calling, and the use of physical punishment. Verbosity refers to the use of lengthy or repetitive verbal responses and reprimands, and expresses a reliance on talking even when talking is ineffective.

The original subscales demonstrated good reliability with alpha coefficients of .83, .82, and .63 and test-retest correlations of 0.83, 0.82, and 0.79 over a two-week period, for the Laxness, Overreactivity, and Verbosity subscales, respectively. In a study of 15 families, subscale scores were significantly correlated with global ratings on observational measures of parenting and children's behaviour (Arnold et al., 1993).

Subsequent factor analyses using PS data from a variety of samples (including parents of adolescents, children in the Head Start program, and children with attention deficit/hyperactivity disorder) have reported strong psychometric support for the Overreactivity and Laxness subscales, but little support for the Verbosity subscale (Collett, Gimpel, Greenson, & Gunderson, 2001; Harvey, Danforth, Ulaszek, & Eberhardt, 2001; Irvine, Biglan, Smolkowski, & Ary, 1999; Reitman et al., 2001). The lack of support for the Verbosity subscale in these studies is consistent with the findings from the original analysis in which the Verbosity factor was clearly identified in nonclinic but not clinic samples.

In each of these later studies the authors reported two-factor solutions corresponding to the Laxness and Overreactivity factors reported by Arnold et al., (1993), but with items that differed from the original scale. It should be noted that discrepancies between the items on the two factors in these studies and the original factor analysis are largely a result of the criteria used for factor loading cutoffs (i.e., factor loadings of .35, .40, and .55). If a consistent factor loading is used as a common criterion across the studies, the items loading on the Laxness and Overreactivity factors are more similar.

The Laxness and Overreactivity subscales have demonstrated concurrent validity, with scores on these subscales significantly differentiating mothers of children with behaviour problems (e.g., children attending clinics or with attention deficit/hyperactivity disorder) from mothers of children without such problems (Arnold et al., 1993; Harvey et al., 2001). Higher scores on the Overreactivity subscale have also been significantly associated with parent reports of greater marital discord, lower parenting support, lower parenting satisfaction, higher psychopathology, higher parenting stress, higher negative feelings about their children and higher levels of children's behaviour problems (Arnold et al., 1993; Collett et al., 2001; Harvey et al., 2001; Reitman et al., 2001). Higher scores on the Laxness subscale have been associated with parent-reports of greater marital discord, higher levels of children's behaviour problems and lower levels of participation in family activities (Arnold et al., 1993; Collett et al., 2001; Irvine et al., 1999). There are no such findings for the Verbosity subscale (Arnold et al., 1993).

The PS has frequently been used in Australian research (Bor et al., 2002; Cann, Rogers, & Matthews, 2003; Cann, Rogers, & Worley, 2003; Sanders et al., 2000; Zubrick et al., 2002), but to date there has been no systematic evaluation of the scale with an Australian sample. The current paper evaluated the factor structure of the PS in a large Australian sample, consisting of parents of preschool aged children from two States and representing different sample types. The first group of parents were those whose children were attending preschools in a metropolitan region of the State of South Australia. The second group of parents were attending parenting groups in a metropolitan region of the State of Victoria. In addition to providing the largest data set used to assess the factor structure of the PS to date, comparisons of the data from these two samples also made it possible to assess the criterion validity of the scale. The concurrent validity of the measure was examined by investigating the relationship between PS scores and measures of children's behaviour problems.

Method

Participants

Participants in the South Australian sample were part of a larger preschool-based longitudinal study conducted by the Women's and Children's Hospital to examine the aetiology and persistence of children's externalising behaviour problems, The 1,163 participants were mothers of children aged three to five years attending preschools in South Australia in 2000.

Participants in the Victorian sample were 493 mothers of three- to five-year-old children in the northern metropolitan region of Melbourne. These mothers were recruited for participation in a group-based behavioural family intervention program conducted at schools, community-based centres or onsite by staff of the Victorian Parenting Centre (now the Parenting Research Centre) between 1999 and 2004.

The demographic characteristics of the two samples, and for the combined sample, are shown in Table I. The age compositions in the two samples were quite different, with 87% of the South Australian sample including parents of four-year-old children, whereas almost half of the Victorian sample included parents of three-year-old children. These differences reflect the age composition of preschool classes in Australia (Australian Bureau of Statistics, 2002) compared with the Victorian sample which did not target parents of a particular child age group.

Table I. Percentage of participants with certain demographic characteristics for the two samples individually and combined

In the combined sample, 52.4% of the children were males and 47.6% were females. A significantly higher proportion of mothers in the Victorian sample were mothers of boys than in the South Australian sample. The high proportion of male children in the Victorian sample is likely due to the fact that this was a group of parents who were actively seeking help for child behaviour problems, and boys are more likely to have behaviour problems for which parents seek help (Marshall & Watt, 1999).

A higher proportion of mothers in the Victorian sample than in the South Australian sample had a tertiary qualification, and the opposite was true of mothers with a technical education qualification (χ² (1, N = 1,638) = 93.2, p < .001). There were no significant differences in the proportions of different family types or in the proportions of mothers in paid employment between the South Australian and Victorian samples.

The characteristics of the combined sample were compared with Australian data for parents of children aged three to four years (Australian Bureau of Statistics, 2004). The sample differed from the Australian population only on the proportion of mothers in paid employment, with the study sample containing a smaller proportion of mothers who were in paid employment than the national data (44% versus 54%). This discrepancy might reflect time constraints of parents who are in paid employment.

Procedures

Procedures for the South Australian sample

Parents were recruited through 35 preschools randomly selected using SPSS from all of the 228 preschools and child parent centres (preschools based at schools) in the Adelaide metropolitan area in South Australia. Questionnaire booklets were distributed to parents by the preschools, and included questions about their child's behaviour, their parenting practices and some demographic items. Of the 1,779 questionnaires distributed, 1,229 were completed and returned (70% response rate). Of the 1,229 respondents, 1,165 (94.8%) were mothers of the index child in the study. Because of the low proportion of fathers who completed questionnaires, only mothers' responses have been used for the analyses in this paper. A subsample (n = 414) of the mothers who participated in the first stage of this study completed a larger questionnaire booklet, which included the PS, two months later. These mothers were selected on the basis of their stage one scores to provide a wide range of parenting behaviour and child behaviour scores. Data from these mothers will be used to examine the test-retest reliability of the PS.

Procedures for the Victorian sample

The majority of mothers who attended the parenting groups did so in response to advertisements for the program circulated through preschools, childcare centres and maternal and child health services. Less frequent sources of referral were nongovernment organizations, family and friends, and government agencies. The groups were advertised as an opportunity for parents to learn a variety of strategies for dealing with their children's misbehaviour (e.g., noncompliance, tantrums). The parenting program was the Level 4 Positive Parenting Program (Triple P) which is a group-based intervention consisting of eight weekly sessions made up of a combination of 2 hour group sessions and half hour telephone sessions. The intervention is described in detail in Cann, Rogers & Matthews (2003). Mothers completed the PS as part of a battery of questionnaires to be returned before the beginning of the intervention.

Measures

Demographic variables

Demographic variables common to both samples were child age and gender, family type, maternal education and maternal employment status.

Parenting Scale (PS; Arnold et al., 1993)

The 30 items of the PS were used to measure parental discipline practices. The reliability and validity of the subscales of the PS have been described in the introduction. It should be noted that the instructions given for the use of the PS with the South Australian and Victorian samples were slightly different. Piloting of the PS in the South Australian study indicated that some parents were confused about the use of the unlabelled rating scale between the anchored items. More detailed instructions and response labels (i.e., always this way, almost always this way, mostly this way and both ways equally) were added to the measure to improve parents' understanding of the response scale. The Victorian sample completed the original version of the PS without the labelled response scale.

Eyberg Child Behavior Inventory (ECBI; Eyberg & Pincus, 1999)

The ECBI is a 36-item parent report measure of children's disruptive behaviour problems, and was administered only to the Victorian sample. Parents rate their child's behaviour on a seven-point scale according to the intensity of difficult child behaviours (Intensity score). After rating the intensity of the difficult behaviour, parents indicate whether the behaviour is a problem or not (Problem score). The inventory is reported to have good reliability and concurrent validity (Boggs, Eyberg, & Reynolds, 1990; Robinson, Eyberg, & Ross, 1980).

Externalising scale of the Provisional version of the Child Behavior Checklist (1.5 – 5) (Achenbach, 1998; Achenbach & Rescorla, 2000)

The 26 items of the Externalising scale of the Provisional Child Behavior Checklist (1.5 – 5) were used in the South Australian sample to measure the levels of children's externalising behaviour problems. The Externalising Scale contains items about behaviours that are aggressive, antisocial and undercontrolled. Parents rate their child's behaviour in the past 6 months on a 3-point scale from 0 (not true) to 2 (very true or often true). The CBCL 1.5 – 5 has adequate test-retest reliability, moderate cross-informant agreement and good discriminant and construct validity (Achenbach, 1998; Achenbach & Rescorla, 2000).

Results

It should be noted that for the results section of this paper, statistical significance refers to a p-value of less than .01 to account for the number of statistical tests performed.

Factor analysis

The factor structure of the 30 items of the PS was investigated using principal components analysis (PCA) with varimax rotation. Consistent with the approach taken by Harvey et al. (2001), an exploratory method of analysis was chosen rather than a confirmatory approach. Confirmatory analyses are usually conducted when there are strong a priori grounds for the hypothesised factors (Fabrigar, Wegener, MacCallum, & Strahan, 1999). In this case, there were a number of reasons that made strong a priori assumptions about the factors unjustified. First, previous studies have all failed to find the original three factors proposed by Arnold et al. (1993), and previous research had given conflicting evidence about the relationships between the factors and items that could be expected (Worthington & Whittaker, 2006). Second, the previous studies were conducted in the United States typically using smaller sample sizes, whereas the current study provides information about the initial validation of the Parenting Scale using a large sample of Australian mothers (Worthington & Whittaker, 2006).

Initially principal components analyses were performed separately for the South Australian and Victorian data. Scree plots from both analyses suggested a three factor solution, and Cattell et al.'s (1969) salient variable similarity index (s) was used to calculate the degree of similarity between the factors for the two samples.1 The values of s were Factor 1 s = .80, Factor 2 s = .84, and Factor 3 s = .58. On the basis of these significant salient variable similarity indices, a decision was made to combine the two samples into a single data set. (N = 1,656). The principal components analysis with varimax rotation was then repeated using this combined sample. The initial principal components analysis of the 30 items resulted in eight factors with eigenvalues greater than one. However, using Cattell's (1966) scree test, a three factor solution was supported. Restricting the analysis to three factors resulted in a clear, interpretable factor structure. The three rotated factors accounted for 15.4%, 13.4%, and 6.4% of the variance, respectively, making a total of 35.2%. The item loadings on the factors are presented in Table II.

Table II. Factor loadings of the parenting scale items in the current study

In this study, items which were selected to contribute to the subscale scores were those with a factor loading cutoff of .35, to replicate the original study by Arnold et al. (1993). As can be seen in Table II, the first factor in the analysis contained all items from the Laxness factor in the original Arnold et al. report, plus Item 1 which in the original scale development did not load on any factor. The second factor contained all of the items from the original Overreactivity factor. The third factor contained all items from the original Verbosity factor, plus items 27 and 20. Items 13 and 5 did not load on any of the three factors. The direction of loading of all items remained the same as for the Arnold et al. (1993) analysis. In summary the items for each factor found in the current analysis were:

Laxness: 1, 7, 8, 12, 15, 16, 19*, 20*, 21, 24, 26*, 30*

Overreactivity: 3*, 6*, 9*, 10*, 14*, 17*, 18, 22*, 25, 28

Verbosity: 2*, 4, 11, 20*, 23*, 27*, 29

(*indicates the item should be reversed when calculating the subscale score). Scale scores are calculated as the mean of the subscale items.

These analyses were replicated for parents of girls and parents of boys, but the significant salient variable similarity indices for the three factors indicated very few differences in the factors obtained in these two analyses: Factor 1, s = .83; Factor 2, s = .96; and Factor 3, s = .64. Therefore, the remainder of the analyses in this paper will be conducted based on the factor analysis for the entire sample.

Reliability

Internal consistency of the PS scores

Reliability estimates for the PS were calculated for the three factors found in the current study using alpha measures of internal consistency. Alpha coefficients were .85 for the Laxness subscale, .81 for the Overreactivity subscale, and .44 for the Verbosity subscale. The PS total score had an alpha coefficient of .84.

Test-retest reliability of the PS scores

The test-retest reliability of the PS scores was examined using scores from the 414 parents from the larger South Australian sample who completed the PS at two assessments approximately two months apart. Pearson's correlations examining the association between parents' scores at the two assessments were large and statistically significant: Total score r = .79; Laxness r = .76; Overreactivity r = .72; and Verbosity r = .63. In addition to correlations, related samples t tests were performed to see if the mean difference between parents' scores at the two assessments was significantly different from zero. Only parents' Verbosity scores were significantly different at the two assessments at the p < .01 level: Verbosity mean (SD) at Time 1 = 4.03 (.69) versus Time 2 = 3.93 (.69), mean difference (SD) = .10 (.59), t(407) = 3.33, p = .001. It is unknown to what extent this small difference might have any clinical relevance.

Concordance with factor structures obtained in previous studies

Cattell et al.'s (1969) salient variable similarity indices (s) were used to compare factors obtained in the current study and previous factor analyses using the PS (Table III). Comparisons with the original PS study by Arnold et al. (1993) could not be included because the authors did not report the factor loadings of all items. High similarity indices (≥.80) were obtained for comparisons between the Laxness and Overreactivity factors obtained in the current study and the corresponding factors in the studies by Irvine et al. (1999) and Collett et al. (2001). The Laxness and Overreactivity factors obtained in the studies by Reitman et al. (2001) and Harvey et al. (2001) yielded similarity indices with the current study of .67 or lower. These two studies also yielded similarity indices with the other factor analytic studies that were generally in the low to moderate range. The lower indices may reflect the smaller sample sizes in these two studies (i.e., less than 200 respondents) which may compromise the reliability of the factor analysis (Tabachnik & Fidell, 1996).

Table III. Salient similarity indices (s) for factors obtained in previous studies and the current analysis (laxness above the diagonal, overreactivity below the diagonal)

	Irvine et al., 1999	Reitman et al., 2001	Harvey et al., 2001	Collett et al., 2001	Current study
Irvine et al., 1999		.61**	.70**	.83**	.87**
Reitman et al., 2001	.45*		.46*	.56**	.67**
Harvey et al., 2001	.61**	.55**		.85**	.67**
Collett et al., 2001	.70**	.30	.73**		.80**
Current study	.82**	.45*	.67**	.86**
*p < .01; **p < .001; n = 30, hyperplane count = 60%+.

Means and standard deviations for the PS subscales by sample characteristics

Mean scores and standard deviations for the PS subscales for the entire sample are presented in Table IV. Using independent samples t-tests, no significant differences were found between the PS scores of parents of boys and parents of girls.

Table IV. Parenting Scale scores by sample characteristics. Means (SD) are given for the total sample and the two subsamples. Pearson's correlation coefficients (r) are given for the measures of children's behaviour problems

The significant correlations between the scores on the three subscales were Laxness and Overreactivity r = .44, Laxness and Verbosity r = .38, and Overreactivity and Verbosity r = .24. The correlations obtained in this study are comparable to those obtained in the original study by Arnold et al. (1993) which ranged from r = 0.35 – 0.43.

Table IV also includes the mean subscale scores for parents from the two subsamples. To test the criterion validity of the PS, independent samples t tests were performed to compare the mean scores of the two subsamples. Parents of children in the Victorian sample (parents attending parenting intervention groups) reported significantly higher scores than parents in the South Australian sample (recruited through preschools) on the PS total score, t(1633) = 12.75, Laxness, t(1647) = 8.39, and Overreactivity, t(1632) = 16.48, subscales. There were no significant differences between the two groups on the Verbosity subscale.

Correlations between PS scores and child behaviour scores

To examine the concurrent validity of the PS scores, correlations were performed between each of the PS subscales and measures of children's behaviour problems (Table IV).

Parents' scores on the Overreactivity and Total Score scales of the PS demonstrated medium-sized correlations with scores on both ECBI subscales and the CBCL Externalising scale score. In general, parents who reported using overreactive techniques more often also reported higher levels of children's behaviour problems. Scores on the Laxness subscale demonstrated small, but significant correlations with the child behaviour measures. Scores on the Verbosity subscale also had small significant correlations with the ECBI Problems subscale and the CBCL Externalising scale scores, but not with the ECBI Intensity scale.

Discussion

This study investigated the psychometric properties of the PS in a sample of Australian parents recruited through preschools and parenting programs. Although the parents in the two subsamples displayed different demographic characteristics (given the different recruitment strategies), as a whole, they had characteristics quite similar to the Australian population from which they were selected (with the only difference being the proportion of mothers in paid employment). The results of the principal components analysis in this study identified a similar factor structure to that originally reported by Arnold and colleagues (1993). The consistency of the results obtained between parents in the two samples and between parents of boys and parents of girls in this study gives further support to the construct validity of the factor structure.

As with previous factor analytic studies using samples of parents from the United States, the Laxness and Overreactivity factors demonstrated high internal consistency and test-retest reliability. The Overreactivity factor also demonstrated good concurrent validity, based on significant associations with children's behaviour problems. The associations between scores on the Laxness subscale and children's behaviour were statistically significant, but modest in size. Scores on both subscales differentiated parents recruited through preschools and those attending parenting programs for their children's behaviour. The differences in the strength of the associations between the Overreactivity and Laxness subscales and measures of children's behaviour in this study are consistent with previous research (Arnold et al., 1993; Collett et al., 2001; Harvey Arnold & O'Leary, 1995, 1997; Harvey et al., 2001; Reitman et al., 2001). This could suggest that the Laxness factor has lower concurrent validity than the Overreactivity factor, but it might also suggest that different family factors to those measured in this and previous studies (e.g., substance use, physical illness, stress) should be identified as determinants or outcomes of lax parenting practices.

The Verbosity factor had lower internal consistency and test-retest reliability than the other two subscales, and consistent with the original factor analysis, the factor structure of the Verbosity subscale was not as strongly supported in different samples. Support for this subscale has only previously been reported in the original study by Arnold et al. (1993). In the current study, the salient similarity indices for the Verbosity factors between parents recruited through preschools and parenting programs, and between parents of boys and girls were lower for the Verbosity subscale than for the Overreactivity and Laxness subscales. Closer inspection of the factors in these samples revealed that all items from the original Verbosity factor loaded on the third factor in the preschool sample but only some items loaded on this factor in the sample of parents attending parenting programs. The same was true for parents of boys compared with parents of girls. These findings duplicate those of Arnold et al. (1993). Further, mothers' scores on the Verbosity subscale did not differentiate parents who were attending parenting programs from parents recruited through preschools, and demonstrated only weak correlations with measures of child behaviour and parent factors. Mean scores on the Verbosity subscale were much higher than mean scores on the other two subscales (and than the scores in the original paper by Arnold et al.), suggesting that these parenting behaviours may be more normative in the Australian population. On the basis of these findings, the Verbosity subscale should be used with caution as a measure of ineffective parenting behaviour. Further investigation of the validity of this subscale is warranted.

These results support the use of the PS as a screening tool to measure overreactive and lax parenting styles with Australian mothers. However the results should be interpreted with some limitations in mind. First, the correlations between the scores on the PS subscales and other measures may be inflated as a result of common method variance. Because the same method (i.e., parent report questionnaires) was used to assess parenting practices and child behaviour, the correlations between these measures may to some extent reflect systematic biases in parents' responses (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). Further research should examine the validity of the subscales using a multimethod multi-informant approach. Second, the self-report nature of the study may have prevented parents from non-English speaking backgrounds from responding. It is possible that the factor structure of the PS might be different for parents from different cultural backgrounds, therefore it is important that the results of this study not be generalised to all mothers. Third, as with many studies of parenting behaviour, this study included only mothers. Only a limited amount of data from fathers was available in both the South Australian and the Victorian samples. Research with a representative sample of fathers is necessary to make sure the PS is a valid and reliable measure of the parenting techniques of fathers. Finally, the fact that mothers in the South Australian sample completed an adapted version of the Parenting Scale (with instructions and response labels added) and the Victorian participants completed the original version of the scale, may have affected the comparability of the scale across the samples. There is limited evidence of this, as the factor analyses performed using the separate sets of data had high similarity indices. It is likely that the use of the amended scale only served to reduce the amount of missing data in the South Australian sample by increasing the ease of interpretability of the response scale. Therefore, future studies using the Parenting Scale might similarly benefit by including similar instructions and response labels.

Acknowledgements

The South Australian component of this research was part of a larger study supported by grants from the Australian Rotary Health Research Fund and the Australian National Health and Medical Research Council. The Victorian component of this work was funded as part of a service delivery agreement between the Victorian Parenting Centre (now the Parenting Research Centre) and the Victorian Department of Human Services.

Notes

1 To calculate s it is necessary to categorise factor loadings as positive salient, negative salient or hyperplane on the basis of an arbitrary criterion factor loading. Consistent with previous research (Livingston, Jennings, Reynolds, & Gray, 2003), the criterion factor loading used to determine the salience of an item in this paper is the ipsative mean plus or minus the standard deviation of the factor loadings. Items with positive salience are those which have a factor loading greater than or equal to the positive value of the criterion factor loading (e.g., ≥0.40). Items with negative salience are those which have a factor loading equal to or lower than the negative value (e.g., ≤−0.40). Hyperplane items are those which are neither positively or negatively salient.

	The coefficient s is then calculated as:
	Where: Table Factor 2 Factor 1 Positive salient Hyperplane Negative salient Positive salient f11 f12 f13 Hyperplane f21 f22 f23 Negative salient f31 f32 f33