Health Science Journal

Keywords

Resilience; Health; Mental health; Mental disorders; Clinical psychology; Questionnaire; Short scale; Screening tool

Introduction

Resilience is the ability to adequately deal with stress and adapt in a functional manner [1,2]. This includes physical stress – such as injuries or illnesses - as well as psychological stress, which can stem from any major or minor life event, e.g. the loss of a loved one, unemployment, a family crisis or similar [3]. Therefore, a resilient individual can cope with adverse circumstances in healthier ways than someone with low resilience, because of their more positive and optimistic selfconcept and outlook on life. In this light, resilience is also a very influential protective factor against mental disorders [4] and a common goal in their treatment [5]. Resilience is associated with self-esteem and self-efficacy [6,7]. Furthermore, life satisfaction is a closely-related construct [8-10]. As mentioned above, resilience can protect individuals against mental disorders and is thus associated with well-being and measures of mental health [11-13]. This means, assessment, practice, as well as maintenance of resilience can be important topics in the treatment of mental disorders [14].

The Resilience Scale (RS) was developed by Wagnild and Young [15]. It is generally considered to be the most reliable and valid instrument for measuring an individual’s resilience [16-18]. Namely, it is valid towards psychological constructs such as hopelessness, social connectedness, life satisfaction, anxiety, depression, stress, and health promoting activities [16,19,20]. Von Eisenhart Rothe et al. [1] developed the Resilience Scale-5 (RS-5) for application in settings that do not allow for long questionnaires, such as large cohort studies and specifically older populations. They found a uni-factorial model consisting of the RS-11 items 3, 6, 7, 8, and 9 (or C, F, G, H, and I) to display the best fit, explaining 57% of total variance.

The main objective of the present study was to validate the RS-5 in a representative sample of the general population. This includes the analysis of item characteristics, validity, reliability, and factor structure, as well as an analysis of measurement invariance, especially across different age groups. Furthermore, differences in resilience based on sociodemographic variables were investigated. Finally, norm values were to be established.

Methods

Sample

The study sample was acquired with the assistance of a demographic research institute (USUMA, Berlin), abiding by the German law of data protection (§30a BDSG, German law of protection of data privacy). Furthermore, the study conformed to the guidelines outlined in the Declaration of Helsinki [21]. The sampling procedure first targeted random sample point areas, then a random household within those areas, and finally chose a person within these households. Of the 8106 potential participants, which had been randomly selected, 3070 (37.9%) either did not respond or refused to take part in the study. All participants gave their informed consent before participating in the study. Where applicable, a caretaker or legal guardian gave consent instead of the participant. The total sample consisted of n=5036 participants, who were at least 14 years of age and living in Germany in 2006. Detailed sample characteristics are displayed in Table 1. The sample can be considered representative of the German population by the criteria of age, sex, and educational level based on official statistics [22]. Sixty-four participants (1.3%) did not complete all items relevant to the RS-5. Those participants differed significantly from the main sample in terms of their age distribution (U=129010.5, n1=4972, n2=64, p=.009) and were excluded from any further analysis leading to a final sample size of n=4972. Participants answered the questionnaires listed below among others in Table 1.

Measures

The Resilience Scale-11 (RS-11; [23,24]) was used to measure resilience. Answer options range from 1 (“strongly disagree”) to 7 (“strongly agree”), and a scale score can be obtained by calculating the sum of the items in question. Von Eisenhart Rothe et al. [1] found a Cronbach’s α of .80 for the shorter RS-5 scale.

To measure symptoms of depression, the Patient Health Questionnaire-2 (PHQ-2; [25,26]) was employed. The two items of the scale, which range from 0 (“not at all”) to 3 (“nearly every day”), are added up to obtain the scale score. Retest-reliability is reported as r=.83.

The General Anxiety Disorder Scale-7 (GAD-7; [27,28]) was utilized to assess the anxiousness of participants. It consists of seven items, which are answered on a 4-point Likert scale, ranging from 0 (“Not at all”) to 3 (“Almost every day”). The sum score can therefore range from zero to 21. Participants with a score above or equal to 10 are generally classified as anxious according to Kroenke, Spitzer, Williams, Monahan, and Löwe [29]. Cronbach’s α is reported as .89 [30].

	n	%	RS-5
			M (SD)
Gender
Female	2,670	53.7	27.00 (5.34)
Male	2,302	46.3	27.42 (5.02)
Age, years (M=48.37; SD=17.98)
<40	1688	34	27.95 (5.09)
40-59	1741	35	27.54 (4.92)
≥60	1543	31	26.00 (5.41)
Education
≤8 years	2,249	45.2	26.02 (5.31)
9 – 11 years	1,701	34.2	27.93 (4.92)
≥12 years	849	17.1	28.79 (4.52)
Student	173	3.5	27.54 (4.87)
Employment status
Working full-time	1793	36.1	28.37 (4.65)
Working part-time	511	10.3	27.61 (4.80)
Retired	1487	30	26.06 (5.39)
Unemployed/working <15 h/week	753	15.1	26.01 (5.37)
Education/training	428	8.6	27.85 (5.28)
Family status
Married	2,664	53.6	27.36 (5.02)
Single	998	20.1	27.61 (5.33)
Committed Relationship	211	4.2	28.84 (4.72)
Separated	61	1.2	26.85 (4.28)
Divorced	471	9.5	27.22 (5.15)
Widowed	567	11.4	25.12 (5.57)
Monthly net household income
<1,500€	1,709	34.4	26.16 (5.48)
1,500€-2,500€	1,937	39	27.20 (4.99)
>2,500€	1,071	21.5	28.85 (4.70)
No answer	255	5.1	27.25 (4.98)

Table 1 Sample characteristics, including RS-5 sum score means and standard deviations.

Self-esteem was measured utilizing the Rosenberg Self- Esteem Scale (RSES; [31-33]), which consists of ten items, half of which are to be inverted before calculating a scale sum score. Participants indicate the extent to which they agree with presented statements on a 4-point scale, ranging from 0 (“do not agree at all”) to 3 (“completely agree”). Internal consistency is reported as α=84.

For the measurement of life satisfaction, the Questionnaire for Life Satisfaction-M (QLS-M; [31-34]), consisting of 16 items was used. Participants first indicate on a 7-point scale how important a given topic is to them, and then specify how satisfied they are with their lives in terms of this topic. A weighted score is obtained by multiplying subjective importance and satisfaction. The sum of all 8 products represents the scale score. Cronbach’s α is reported as .83 by Daig et al. [34].

The screening version of the Recalled Parental Rearing Behavior Questionnaire [35] was utilized to assess the upbringing of participants. It consists of six items for two scales, asking for memories of the participant’s father and mother, respectively, on a 4-point scale, ranging from 1 (“No, never”) to 4 (“Yes, always”). Two items each make up one of three latent factors, which are “Rejection and Punishment” (RP), “Emotional Warmth” (EW), and “Control and Overprotection” (CO). No measures of reliability are reported because of the measure’s brevity.

Statistical analysis

IBM SPSS Statistics 23 was used for most statistical calculations. The confirmatory factor analysis (CFA) was conducted in IBM AMOS 23. The Pearson product-moment correlation coefficient was used for all correlations. The α level of significance tests was .05 unless noted otherwise. Means and standard deviations of the RS-5 scale and its items were determined, in addition to item difficulty and corrected itemtotal correlations. The RS-5 scale and items were tested for normality of distribution by calculating skewness and kurtosis. For the CFA, covariance matrices and the maximum likelihood method were utilized. Common fit indices were employed to judge model fit, including χ², the minimum discrepancy divided by degrees of freedom (CMIN/DF), the comparative fit index (CFI), the Tucker–Lewis Index (TLI), the standardized root mean square residual (SRMR), and the root mean square error of approximation (RMSEA) and its 90% confidence interval. Recommended levels for these measures are reported as lower than five for CMIN/DF, larger than .95 for CFI and TLI, lower than .08 for SRMR and RMSEA [36-38].

Measurement invariance was tested in three steps using multiple-group analysis. First, the configural model – without any constraints – was compared to the metric model, which constrains unstandardized item loadings to be equal across groups. Secondly, the metric model and the scalar model, which constrains unstandardized item loadings and intercepts across groups, were compared. Commonly used fit indices for these comparisons are the difference in CFI and gamma hat [39,40]. The χ²-statistic is also considered and reported despite its sensitivity to sample size.

Analyses of variance were conducted in order to test for differences in RS-5 scores across sociodemographic groups of gender, age, and education level. Additionally, post hoc tests in the form of Tukey’s HSD were conducted. Counteracting the accumulation of α error probability, a significance level of .01 was employed in both the ANOVAs and the post hoc tests. Cohen’s d is also reported, with greater than .2 being a small, greater than .5 being a medium, and greater than .8 being a large effect [41].

Results

Reliability and item characteristics

Internal consistency of the RS-5 scale was α=.87. Skewness and Kurtosis are within the commonly agreed upon thresholds of lower than 1 for skewness and lower than 3 for kurtosis, indicating a normal distribution of the RS-5 items and scale [42,43]. The difficulty indices of RS-5 items were between .70 and .76, which means the items were generally answered in the affirmative by most participants. Furthermore, the items satisfied the common cutoff point for corrected item-total correlations of being higher than .50 [44]. Details can be found in Table 2.

	M(SD)	Skewness	Kurtosis	P	rit
RS C	5.56 (1.30)	-0.83	0.34	0.76	0.71
RS F	5.46 (1.29)	-0.73	0.19	0.74	0.67
RS G	5.54 (1.25)	-0.8	0.42	0.76	0.78
RS H	5.43 (1.29)	-0.68	0.08	0.74	0.62
RS I	5.22 (1.32)	-0.51	-0.16	0.7	0.65
RS-5 scale	27.20 (5.20)	-0.62	0.34
P=item difficulty index; rit=corrected item-total correlation

Table 2: Descriptive item characteristics.

Factor structure

Von Eisenhart Rothe et al. [1] proposed a model consisting of the RS-items C, F, G, H, and I. In the present study, the CFA of those same items shows acceptable to good model fit for the entire population, as can be seen in Table 3.

	χ2(df)	CMIN/DF	CFI	TLI	SRMR	RMSEA
RS-5 model	258.879 (5)	51.78	0.978	0.96	0.028	0.101
CFI=Comparative-fit index; TLI=Tucker-Lewis-index; SRMR=Standardized root mean square residual RMSEA=Root mean square error of approximation

Table 3 Model fit indices in the CFA.

Loadings of individual items on the latent variable were between .66 and .88. Total variance explained for the five variables was approximately 65%.

The results of the analysis of measurement invariance across sociodemographic groups of gender, age, and education are displayed in Table 4.

Model	χ2(df)	Δ χ2	Δ p	CFI		ΔCFI	GH			ΔGH
Gender
Male	130.849 (5)							1			1
Female	133.431 (5)							1			1
Multigroup analysis
Configural invariance	264.280 (10)							1			1
Metric invariance	290.106 (14)	25.8			< .001			1	0		1	0
Scalar invariance	317.960 (20)	27.9			< .001			1	0		1	0
Age, years
<40	105.135 (5)							1			1
40-59	88.779 (5)							1			1
≥60	70.289 (5)							1			1
Multigroup analysis
Configural invariance	264.203 (15)							1			1
Metric invariance	280.238 (23)	16			0			1	0		1	0
Scalar invariance	354.521 (35)	74.3			< .001			1	0		1	0
Education
≤8 years	101.813 (5)							1			1
9 – 11 years	100.281 (5)							1			1
≥12 years	80.572 (5)							1			1
Student	13.368 (5)							1			1
Multigroup analysis
Configural invariance	296.073 (20)							1			1
Metric invariance	320.516 (32)	24.4			0			1	0		1	0
Scalar invariance	426.452 (50)	106			< .001			1	0		1	0
df=degrees of freedom; CFI=comparative fit index; GH=gamma hat.

Table 4 Fit indices for the multigroup analysis.

The differences in CFI and gamma hat between models did not exceed .01. Therefore, scalar invariance could be shown for males and females as well as for different age and education groups.

Mean differences with regard to sociodemographic variables

Men were found to score significantly higher when compared to women, t(4933.08)=2.86, p=.004, d=.08. Age groups also differed with statistical significance in RS-5 score, F(24969)=64.35, p<.001, η²=.03. So did education groups, F(34968)=80.49, p<.001, η²=.05. Groups of employment status differed in their resilience scores, F(44967)=55.48, p<.001, η²=. 04, as well did groups of family status F(54966)=24.78, p<.001, η²=.02. Finally, participants of different monthly net household income showed significant differences in RS-5 scores, F(34968)=55.48, p<.001, η²=.04. Means and standard deviations for all groups can be found in Table 1.

Validity

To investigate the validity of the RS-5 scale, correlations to related psychological constructs were calculated and are reported in Table 5. A high positive correlation with the RSES was expected and found. Furthermore, there were moderate negative correlations to measures of psychopathology such as the PHQ-2 and the GAD-7. The QLS-M, which measures life satisfaction, was also moderately associated with the RS-5 scale. The FEE subscales were weakly correlated with the RS-5.

	RS-5
RSES	.56*
PHQ-2	-.31*
GAD-7	-.26*
QLS-M	.40*
FEE-RPa	-.14*
FEE-EWb	.13*
FEE-COc	-0.01
* p<.001; a=Rejection and Punishment; b=Emotional Warmth; c=Control and Overprotection

Table 5 Correlations between the RS-5 and related psychological measures.

Discussion

The main objective of the present study was to validate the RS-5 in the general population, specifically younger participants, and examine its psychometric properties. The RS-5 scale was shown to have good internal consistency with an α of .87, which is only slightly lower than that found in Schumacher’s analysis of the longer RS-11 (α=.91) [23]. This is evidence of the RS-5 scale’s capability to measure resilience reliably with just a fifth of the items of the original RS-25 scale. Furthermore, factor loadings for all items of the RS-5 on a common factor were well above .50, indicating the validity of the single factor solution [44]. CFI, TLI, and SRMR demonstrate very good fit, while RMSEA along with the χ²-statistic and the CMIN/DF revealed mediocre to unacceptable fit. Both, the χ²- test and the CMIN/DF, however, are well-known to be sensitive to sample size [45] and should thus be interpreted with caution. Overall, model fit can thus be considered acceptable. Additionally, measurement invariance could be shown across groups of gender, age, and education. This means, the model fits the data equally well for any of these sociodemographic subgroups and thus allows for statistical comparisons between them.

The construct validity of the RS-5 was shown via correlations to several related psychological constructs. Firstly, the scale correlates highly with the RSES. This was expected, as the extent to which an individual can be resilient is closely linked to their self-esteem [6,7]. Furthermore, the expected correlation to the QLS-M, measuring life satisfaction, was moderately high. This finding is consistent with prior research indicating that resilient individuals are often more satisfied with life [8-10]. Secondly, there was a negative association between RS-5 and measures of psychopathology (PHQ-2 and GAD-7). Per definition, resilience describes an individual’s fortitude against obstacles and adverse circumstances, and, therefore, moderate negative correlations to aforementioned measures were anticipated and could be confirmed [11-13]. Finally, the FEE correlated very weakly with the RS-5 questionnaire, indicating that parental rearing behavior is just barely or not at all associated with an individual’s resilience later in life.

The analysis of differences between individuals in resilience based on sociodemographic characteristics revealed effects of varying size that should be more thoroughly investigated in further studies.

Study limitations

Table 6, the present study utilized the 11-item version of the Resilience Scale. Thus, further validation of the RS-5 should be carried out, in order to rule out any external influences on the measurement.

RS-5 Sum Score	Male		Female
	<60 years(n=1,585)	=60 years	<60 years	=60 years
		(n=717)	(n=1,844)	(n=826)
5	0	0	0.2	0.1
6	0	0.1	0.2	0.1
7	0	0.3	0.3	0.4
8	0	0.3	0.4	0.5
9	0	0.3	0.7	0.6
10	0.1	0.4	0.8	0.7
11	0.4	0.8	0.9	1.1
12	0.6	1.1	1	1.3
13	0.8	1.5	1.1	1.9
14	1.1	2	1.4	2.4
15	1.4	2.8	1.7	3.3
16	1.8	3.5	2.2	4.5
17	2.8	4.5	2.9	6.9
18	3.8	5.7	4.1	8.8
19	6.2	9.3	6	13.9
20	9.5	13.4	9	21.1
21	11.4	17.6	12	25.5
22	13.9	22.5	15.5	31
23	18.2	27.2	19.5	36.2
24	23.5	31.9	24.9	42.4
25	29	39.2	31.3	48.2
26	35.5	46.3	37.3	55.9
27	42.8	52	44.6	63.2
28	51.9	59.1	51.7	68
29	61.1	66.7	59.5	74.6
30	69	76.3	68.3	79.5
31	76.6	82.1	74.9	84.6
32	81.1	87.3	81.2	89.1
33	87.2	90.9	86.4	93.3
34	91	93.9	90.2	94.9
35	100	100	100	100

Table 6 Normative percentile values for the general population.

Conclusion

To sum up, the RS-5 is a reliable and valid measure of resilience for older and younger populations alike. It displays a good model fit and can be recommended for usage in research and clinical applications.

11314

References

1. Von EisenhartRothe A, Zenger M, Lacruz ME, Emeny R, Baumert J, et al. (2013) Validation and development of a shorter version of the resilience scale RS-11: results from the population-based KORA–age study. BMC Psychology 1:1-7.
2. Luthar SS, Cicchetti D, Becker B (2000) The Construct of Resilience: A Critical Evaluation and Guidelines for Future Work. Child Dev71: 543-562.
3. Shastri PC (2013) Resilience: Building immunity in psychiatry. Indian J Psychiatry 55:224-234.
4. Kent M, Rivers CT, Wrenn G (2015) Goal-Directed Resilience in Training (GRIT): A Biopsychosocial Model of Self-Regulation, Executive Functions, and Personal Growth (Eudaimonia) in Evocative Contexts of PTSD, Obesity, and Chronic Pain. BehavSci 5:264-304.
5. Schwarzer R, Warner LM (2013) Perceived Self-Efficacy and its Relationship to Resilience. In: Prince-Embury S, Saklofske HD (eds.) Resilience in Children, Adolescents, and Adults: Translating Research into Practice, Springer, New York. 139-150.
6. Veselska Z, Geckova AM, Orosova O, Gajdosova B, Van Dijk JP, et al. (2009) Self-esteem and resilience: the connection with risky behavior among adolescents. Addict Behav 34:287-291.
7. Akbar M, Akram M, Ahmed M, Hussain MS, Lal V, et al. (2014) Relationship between Resilience and Life Satisfaction among Nomadic. Int J InnovAppl Stud 6:515-529.
8. Cohn MA, Fredrickson BL, Brown SL, Mikels JA, Conway AM (2009) Happiness Unpacked: Positive Emotions Increase Life Satisfaction by Building Resilience. Emotion 9:361-368.
9. Shi M, Wang X, Bian Y, Wang L (2015) The mediating role of resilience in the relationship between stress and life satisfaction among Chinese medical students: a cross-sectional study. BMC Med Edu 15:1-7.
10. Macleod S, Musich S, Hawkins K, Alsgaard K, Wicker ER (2016) The impact of resilience among older adults. GeriatrNurs.
11. Mguni N, Bacon N, Brown JF (2012) The wellbeing and resilience paradox. The Young Foundation, London.
12. Wiseman T, Foster K, Curtis K (2016) The experience of emotional wellbeing for patients with physical injury: A qualitative follow-up study. Injury.
13. Padesky CA, Mooney KA (2012) Strengths-Based Cognitive–Behavioural Therapy: A Four-Step Model to Build Resilience. ClinPsycholPsychother19: 283-290.
14. WagnildGM,Young HM (1993) Development and psychometric evaluation of the Resilience Scale. J NursMeas 1:165-178.
15. Ahern NR, Kiehl EM, Sole ML, Byers J (2006) A review of instruments measuring resilience. Issues ComprPediatrNurs 29:103-125.
16. Wagnild G (2009) A review of the Resilience Scale. J NursMeas 17:105-113.
17. Windle G, Bennett KM, Noyes J (2011) A methodological review of resilience measurement scales. Health Qual Life Outcomes 9: 8.
18. Abiola T, Udofia O (2011) Psychometric assessment of the Wagnild and Young's resilience scale in Kano, Nigeria. BMC Res Notes 4: 509.
19. Heilemann MV, Lee K, Kury FS (2003) Psychometric properties of the Spanish version of the Resilience Scale. J NursMeas 11:61-72.
20. World Medical Association (2013) Declaration of Helsinki: ethical principles for medical research involving human subjects. Jama 310:2191-2194.
21. Kocalevent RD, Zenger M, Heinen I, Dwinger S, Decker O, Brähler E (2015) Resilience in the General Population: Standardization of the Resilience Scale (RS-11). PLoS ONE 10:e0140322.
22. Schumacher J, Leppert K, Gunzelmann T, Strauß B, Brähler E (2005) Die Resilienzskala—EinFragebogenzurErfassung der psychischenWiderstandsfähigkeitalsPersonmerkmal. ZeitschriftfürKlinischePsychologie, Psychiatrie und Psychotherapie 53:16–39.
23. Kroenke K, Spitzer RL, Williams JB (2003) The Patient Health Questionnaire-2: validity of a two-item depression screener. Med Care 41:1284-1292.
24. Spitzer RL, Kroenke K, Williams JB (1999) Validation and utility of a self-report version of PRIME-MD: the PHQ primary care study. Primary Care Evaluation of Mental Disorders. Patient Health Questionnaire. Jama 282:1737-1744.
25. Spitzer RL, Kroenke K, Williams JW, Löwe B (2006) A brief measure for assessing generalized anxiety disorder: The gad-7. Archives of Internal Medicine 166:1092-1097.
26. Swinson RP (2006) The GAD-7 scale was accurate for diagnosing generalised anxiety disorder. Evid Based Med 11:184.
27. Kroenke K, Spitzer RL, Williams JB, Monahan PO, Löwe B (2007) Anxiety disorders in primary care: prevalence, impairment, comorbidity, and detection. Ann Intern Med 146:317-325.
28. Lowe B, Decker O, Muller S, Brähler E, Schellberg D, et al. (2008) Validation and standardization of the Generalized Anxiety Disorder Screener (GAD-7) in the general population. Med Care 46:266-274.
29. Rosenberg M (1965) Society and the adolescent self-image. Princeton University Press, Princeton, NJ.
30. Roth M, Decker O, Herzberg PY, Brahler E (2008) Dimensionality and Norms of the Rosenberg Self-esteem Scale in a German General Population Sample. European Journal of Psychological Assessment 24:190-197.
31. Von Collani G, Herzberg PY (2003) EinerevidierteFassung der deutschsprachigenSkalazumSelbstwertgefühl von Rosenberg. ZeitschriftfürDifferentielle und DiagnostischePsychologie 24:9-22.
32. Daig I, Spangenberg L, Henrich G, Herschbach P, Kienast T, et al. (2011) Alters- und geschlechtspezifischeNeunormierung der FragenzurLebenszufriedenheit (FLZM) für die Altersspanne von 14 bis 64 Jahre. ZeitschriftfürKlinischePsychologie und Psychotherapie 40:172-178.
33. Petrowski K, Paul S, Zenger M, Brahler E (2012) An ultra-short screening version of the Recalled Parental Rearing Behavior questionnaire (FEE-US) and its factor structure in a representative German sample. BMC Med Res Methodol 12: 169.
34. Hu L, BentlerPM (1998) Fit Indices in Covariance Structure Modeling: Sensitivity to Underparameterized Model Misspecification. Psychological Methods 3: 424-453.
35. Hu L, BentlerPM (1999) Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling 6: 1-55.
36. Schermelleh-Engel K, Moosbrugger H, Müller H (2003) Evaluating the fit of structural equation models: Tests of significance and descriptive goodness-of-fit measures. Methods of Psychological Research Online 8:23–74. Available from https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.509.4258&rep=rep1&type=pdf
37. Cheung GW, Rensvold RB (2002) Evaluating Goodness-of-Fit Indexes for Testing Measurement Invariance. Structural Equation Modeling 9:233–255.
38. Milfont TL, Fischer R (2010) Testing measurement invariance across groups: Applications in cross-cultural research. International Journal of Psychological Research 3:112-131.
39. Bulmer MG (1979) Principles of Statistics. Dover Publications, Mineola, NY.
40. Byrne BM (2010) Structural Equation Modelling with AMOS: Basic Concepts, Application, and Programming. Taylor & Francis, New York.
41. Hair J, Black W, Babin B, Anderson R (2010) Multivariate Data Analysis. Prentice Hall, Upper Saddle River, NJ.
42. Joereskog KG, Soerbom D (1993) Lisrel 8: Structural equation modeling with the SIMPLIS command language. Erlbaum, Hillsdale, NJ.