Assessing metacognitive beliefs about worry: validation of German versions of the Why Worry Scale II and the Consequences of Worry Scale

Background Metacognitive beliefs have been proposed to play a key role in initiating and maintaining worry. The Why Worry-Scale-II (WW-II) and Consequences of Worry Scale (COWS) are self-report questionnaires assessing positive and negative metacognitive beliefs. The main goal of this study was to validate German versions of these two questionnaires. Method N = 603 participants completed a questionnaire battery, including the two self-report measures of metacognitive beliefs. We conducted confirmatory factor analyses, calculated internal consistencies, and examined convergent and divergent validity. In addition, the questionnaires’ power in predicting worry, repetitive negative thinking (RNT) and generalized anxiety disorder (GAD) symptoms were investigated. Results The factor structure of the original versions could be replicated for both measures. Furthermore, the translated questionnaires demonstrated excellent internal consistency and evidence of convergent and divergent validity. Importantly they also possessed predictive power in explaining worry, RNT and GAD symptoms, even over and above the Metacognitions Questionnaire-30 (MCQ-30) as the current gold standard. Conclusions Overall, our findings suggest that the WW-II and COWS show solid psychometric properties and are useful in measuring metacognitive beliefs independently from the MCQ-30.


INTRODUCTION
Worry is a repetitive thought activity that ranges from an everyday phenomenon to an excessive and seemingly uncontrollable manifestation. Excessive uncontrollable worry is the key characteristic of generalized anxiety disorder (GAD;APA, 2013), while worry also is an everyday phenomenon and most researchers conceptualize it on a continuum, with quantitative rather than qualitative differences between worry in non-clinical vs. clinical samples (Ehring & Watkins, 2008). Current cognitive models of dysfunctional worry and GAD suggest that metacognitive beliefs (''thoughts about thoughts'') are crucial in initiating 1 The MCQ-65 (8 out of 16 items measuring negative beliefs directly refer to the uncontrollability of worrying) and MCQ-30 (three out of six items aim at capturing the perceived lack of control: ''My worrying thoughts persist, no matter how I try to stop them'', ''When I start worrying I cannot stop'', ''I cannot ignore my worrying thoughts'') show overlap with the PSWQ (example item ''Once I start worrying, I cannot stop.'') that is often used as a criterion. and maintaining worrisome thinking (Behar et al., 2009) and play an important role across psychopathologies (Sun, Zhu & So, 2017).
In his Metacognitive Model, Wells (1995) suggests that positive metacognitions (e.g., ''Worry helps to cope.'') initiate worry in response to internal or external cues (e.g., ''I might fail the test.''). Worrying then activates negative metacognitions (e.g., ''Worry is uncontrollable.''), leading to meta-worry or 'worry about worry' (e.g., ''If I start worrying now, I will not be able to stop.''), which again evokes engagement in thought control. Paradoxically, thought suppression leads to the maintenance of worry, strengthens the negative metacognitive belief and causes negative emotional responses. Thus, evidence shows that worry maintains the problem instead of reducing it while having an alarming rather than calming effect.
Likewise, the Intolerance of Uncertainty Model (Dugas et al., 1998) presents positive metacognitions as a key construct. Individuals with high intolerance of uncertainty are proposed to experience stress in response to ambiguous situational cues, leading to the activation of positive beliefs about worry (i.e., beliefs in worry as an aid to problem solving and motivation, as protection from negative emotions, as a positive personality trait, and as a possibility to directly alter events; Francis & Dugas, 2004), whereby worrisome thinking is initiated.
Most studies to date have used the Metacognitions Questionnaire (MCQ; Cartwright-Hatton & Wells, 1997; short form MCQ-30 Wells & Cartwright-Hatton, 2004), which is often regarded as the gold standard measure of metacognitive beliefs about worry and is a well-validated measure (Wells & Cartwright-Hatton, 2004). However, the MCQ-30 can be criticized on at least two grounds: first, it assesses positive and negative metacognitive beliefs about worry unidimensionally and does not capture subdimensions of them (e.g., subscales of positive metacognitions as postulated by Dugas & Robichaud, 2012). In addition, it has been argued that predicting excessive worrying by the negative metacognition-subscale referring to uncontrollability and danger may be circular, as there is a conceptual overlap between the predictor (i.e., the belief that worry is uncontrollable) and the criterion (i.e., self-report of uncontrollable worry) 1 (Behar et al., 2009;Gerlach, Andor & Patzelt, 2008;Wells, 2005).
Due to these limitations, it appears necessary to additionally use questionnaire measures beyond the MCQ-30 when investigating the relationship between metacognitive beliefs and worry. The Why Worry Scale II (WW : Freeston et al., 1994;WW-II ;Gosselin et al., 2003) and the Consequences of Worry Scale (COWS; Davey, Tallis & Capuzzo, 1996) are alternative questionnaire measures assessing metacognitions about worry. The WW-II 2 In contrast to the MCQ, the COWS directly refers to uncontrollability with only one out of 17 items (''Deep down I know I do not need to worry that much but I can't help it.'').
specifies positive beliefs about worry on five subscales, while the COWS is comprised of items referring to both negative (three subscales) as well as positive consequences (two subscales) about worrying. Importantly, these measures can be expected to overcome some of the problems related to an exclusive use of the MCQ-30: first, both questionnaires examine different facets of positive and negative metacognitions. In addition, while the MCQ-30 subscale measuring negative beliefs about worry is mainly focused on the uncontrollability of worry, the COWS assesses a broad range of negative metacognitions, including worrying disrupting effective performance, exaggerating the problem and causing emotional discomfort. 2 Therefore, it can be argued that the questionnaire shows less criterion contamination with measures of excessive worrying than the MCQ-30.
This study aimed at validating German versions of the WW-II and the COWS. The questionnaires were translated into German, followed by a test of their factor structure and a psychometric evaluation. We expected to find the proposed five-factor structures for the translated versions as well as evidence for good psychometric attributes that have been reported for the original questionnaires (Covin, Dozois & Westra, 2008;Hebert et al., 2014). Additionally, we expected a diagnostic value in the sense of variance explanation in worry and related phenomena based on the WW-II and COWS-also over and above the MCQ-30.

METHOD Participants
Participants for the validation study were invited via the online-panel PsyWeb (https://psyweb.uni-muenster.de/). Inclusion criteria were German as the first language and at least 18 years of age.
Eight hundred eighty panel members started the study, n = 274 did not complete the study and n = 3 withdrew their consent for their data being analyzed at the end of the study. Therefore, 603 participants (69% female) between 18 and 84 years of age (M = 46.40, SD = 13.94) were included in the analyses (education level reported most frequently: 64% had completed an academic type of high school; employment status reported most frequently: 61% classified themselves as currently employed).
The average PSWQ score (M = 40.66, SD = 7.86: below the cut-off score 62 for nonclinical samples; Behar et al., 2003) and the average GAD-Q-IV score (M = 3.78, SD = 3.77: below the cut-off score of 7.67; Moore et al., 2014) overall reflect the non-clinical nature of the sample.
The Perseverative Thinking Questionnaire (PTQ; Ehring et al., 2011) aims at capturing the process of recurring thinking. Independent of a specific type of mental illness, it focuses 3 A third scale named ''negative beliefs about the need to control thoughts'' captures metacognitive beliefs about the coping strategy of thought control, a phenomenon that in the Metacognitive Model (Wells, 1997;Wells, 2005) is conceptually separated from metacognitive beliefs that play a role in initiating type 1 and type 2 worry. Two other subscales measure ''cognitive confidence'' and ''cognitive selfconsciousness''. on repetitive negative thinking (RNT) using 15 items with regard to the key characteristics of the phenomenon (repetitiveness; intrusiveness; difficulties with disengagement) and two further features (unproductiveness of RNT; mental capacity captured by RNT) that have to be rated on a five-point Likert scale ranging from ''never'' to ''always''. The measure has been shown to possess good psychometric properties (Ehring et al., 2011; in this sample α = .96).
The Generalized Anxiety Disorder Questionnaire-IV (GAD-Q-IV; Newman et al., 2002) examines the DSM-IV criteria of GAD with nine items measuring excessiveness and uncontrollability of worry (German version: Hoyer). The questionnaire consists of five dichotomous items assessing the intensity, perceived uncontrollability and specific triggers of worries, a list of the most frequent worry topics, while afterwards the presence or absence of six related physical symptoms has to be reported and ratings of functional impairment and subjective distress (nine-point Likert scales ranging from ''none'' to ''very severe'') are required. High test-retest reliability, diagnostic sensitivity and specificity, as well as convergent and discriminant validity have been reported (Newman et al., 2002;Robinson, Klenck & Norton, 2010). Moore et al. (2014) present a total score of 7.67 as an appropriate cut-off score to screen for GAD in primary care.
The  Arndt et al., 2011a;Arndt et al., 2011b) uses 65 items (original version), respectively 30 items (short form) to measure metacognitive beliefs about worry (example items: ''My worrying is dangerous for me.'', ''Worrying helps me to solve problems.''). We used the MCQ-30, which comprises five scales, of which two specifically refer to negative and positive metacognitive beliefs initiating type 1 and type 2 worry (negative beliefs about uncontrollability and danger of worrisome thoughts; positive beliefs about worry). 3 It has been shown to possess good internal consistency and convergent validity as well as good test-retest reliability (Arndt et al., 2011a;Arndt et al., 2011b;Möbius & Hoyer, 2003;Wells & Cartwright-Hatton, 2004; in this sample α = .88 for all 30 items).
The Why Worry Scale II (WW: Freeston et al., 1994;WW-II;Gosselin et al., 2003) is a 25-item self-report questionnaire to measure positive beliefs about worry (e.g., ''If I did not worry, I would be careless and irresponsible.'') on five subscales (worry facilitates problem solving; worry enhances motivation; worry protects against negative emotions; worry prevents negative outcomes; worry reflects a positive personality trait). Items have to be rated on a five-point Likert scale ranging from ''not at all true'' to ''absolutely true''. Excellent internal consistency and good test-retest reliability as well as a predictive power for worry severity could be demonstrated for the English version (Hebert et al., 2014). With permission from the original authors of the measure, the questionnaire was translated into German by the authors of this study.
The Consequences of Worry Scale (COWS; Davey, Tallis & Capuzzo, 1996) comprises 29 items capturing negative beliefs (e.g., ''Worry gets me worked up.'') and positive beliefs (e.g., ''Worrying stimulates me.'') about the consequences of worrying. The items regress on two subscales (negative beliefs, positive beliefs) and five factors (worrying disrupts effective performance, worrying exaggerates the problem, worrying causes emotional 4 Examining modification indices for the CFA revealed that COWS item 8 shows a similar amount of loading on factor II as it does regarding factor I, so it might not be possible to unambiguously ascribe it to one or the other subscale and one might consider deleting it when using the questionnaire in future research if that is seen as problematic. discomfort, worry motivates, worry helps analytic thinking). Responses are rated on a five-point Likert scale. Good internal reliability could be reported for the subscales of the original version (Covin, Dozois & Westra, 2008;Davey, Tallis & Capuzzo, 1996). The questionnaire was translated into German by the authors of this study with permission of the original author.

Procedure
The German versions were generated following standard procedures (Schmitt & Eid, 2007). The questionnaires were first translated from English to German by the authors. Next, a bilingual psychologist who was unaware of the original questionnaires back-translated the material. In case of inconsistencies, a consensus-finding process was initiated. Afterwards a group of psychology students checked the translated versions for comprehensibility. Participants of the validation study completed all measures online. Before completing the questionnaires, written informed consent was acquired.

Ethics approval and consent to participate
All procedures were performed in compliance with relevant laws and institutional guidelines following the Helsinki Declaration of 1975, as revised in 2013. We observed privacy rights, obtained written informed consent from all individual participants included in the study and disclosed any conflicts of interests with study participants.
In order to examine the factor structure of the questionnaires, confirmatory factor analyses with diagonally weighted least squares (DWLS) estimation (Mindrila, 2010) were conducted using R Statistical Software (R Development Core Team, 2008).

Confirmatory factor analyses
Two confirmatory factor analyses (CFA) were conducted to test the factor structures of the WW-II and COWS. On the basis of the theoretical and empirical literature, more precisely, based on the results of the factor analyses in the original papers (Davey, Tallis & Capuzzo, 1996;Gosselin et al., 2003), five-factor structures were assumed for both questionnaires. With factor loadings greater than 0.35 (see Tables 1 and 2; c.f., Field, 2007), all items loaded significantly on the corresponding factor. 4 The individual CFAs altogether revealed a good fit to the structure. When modeling large sample sizes, the chi-square test is considered very strict, so the unwanted significant results (WW-II: chi-square = 666.428; df = 265; p < .001; COWS: chi-square = 1363.271; df = 367; p < .001) seem tolerable and alternative measures of approximate fit were brought in. The RMSEA (WW-II: RMSEA = .050; COWS: RMSEA = .067) provides a good fit, as both values do not exceed the suggested maximum of .07 (Steiger, 2007). As for the SRMR, values as high as .08 are deemed acceptable (Hu & Bentler, 1999), which was slightly exceeded by the COWS (SRMR = .063 for the Why Worry-II and .087 for

Notes.
Factor loading are significant with p < 0.001. Subscales WW-II: I worry facilitates problem solving, II worry enhances motivation, III worry protects against negative emotions, IV worry prevents negative outcomes, V worry reflects a positive personality trait.

Convergent and divergent validity
Inter-correlations were calculated for the WW-II/COWS subscales (Table 4) and for all measures used in this study (Table 5).

Divergence
As hypothesized, subscales measuring positive metacognitions showed generally low and mostly non-significant associations with those capturing negative beliefs (−.26 < r's < .21). With regard to MCQ-neg, WW-II total score (r = .27) and COWS-pos (r = .14) likewise revealed small correlations. The same applies for COWS-neg and the relationship with MCQ-pos (r = −.15).

Notes.
Factor loading are significant with p < 0.001. Subscales COWS: COWS-negative: I worrying disrupts effective performance, II worrying exaggerates the problem, III worrying causes emotional discomfort; COWS-positive: IV worry motivates, V worry helps analytic thinking.

WWII and COWS predicting worry, RNT and GAD symptoms
To test whether positive and negative metacognitions substantially contribute to predicting worry, repetitive negative thinking, and GAD symptoms, respectively, stepwise multiple regression analyses were conducted with PSWQ, PTQ, and GAD-Q-IV scores as dependent variables and the Why Worry-II and COWS subscales as predictor variables. In Step 1, all five subscales of the WW-II assessing positive metacognitions were included. In Step 2, the COWS subscales measuring positive metacognitions (COWS-pos) were entered. In a final Step 3, the COWS subscales referring to negative beliefs (COWS-neg) were included as additional predictors. To check for multicollinearity, all variables were inspected with regard to their specific tolerance level (which for all variables did not show below the cut-off .25 recommended by Urban & Mayerl, 2006) and Variance Inflation Factor (VIF; which for all variables did not show above the cut-off 5.0 recommended by Urban & Mayerl, 2006).
As displayed in Table 6, of all subscales measuring positive metacognitions, Why Worry-II scale III (''worry protects against negative emotions'') showed the closest relationship

Notes.
Subscales WW-II: I worry facilitates problem solving, II worry enhances motivation, III worry protects against negative emotions, IV worry prevents negative outcomes, V worry reflects a positive personality trait. COWS-negative: I worrying disrupts effective performance, II worrying exaggerates the problem, III worrying causes emotional discomfort. COWS-positive: IV worry motivates, V worry helps analytic thinking; for means and standard deviations see Table 5.
When focusing on repetitive negative thinking (PTQ) as the dependent variable, COWS subscale IV (''worry helps analytic thinking'') additionally made a unique contribution to variance explanation (Step 3: β = .10, p = .03). As for the GAD-Q-IV, Why Worry-II factor V (''worry reflects a positive personality trait'') additionally contributed to the prediction of GAD symptoms at a statistically significant level (step 3: β = .16, p = < .01).

MCQ-30, WW-II and COWS predicting worry, RNT and GAD symptoms
In a final set of regression analyses, we tested whether WW-II and COWS still predicted worry, repetitive negative thinking and GAD symptoms, when the MCQ-30 as the gold standard measure of metacognitions is controlled.
Predicting worry (PSWQ): When adding MCQ-pos (Table 7, Step 1) and afterwards the WW-II scales as well as both COWS subscales referring to positive beliefs about worry, the prediction is optimized (Step 2), while it can be further improved by also using MCQ-neg (Step 3). However, when entering all three COWS subscales referring to negative beliefs about worry, variance explanation is not improved (Step 4).  Notes. * p < .05, * * p < .01, * * * p < .001 (two-sided). Subscales WW-II: I worry facilitates problem solving, II worry enhances motivation, III worry protects against negative emotions, IV worry prevents negative outcomes, V worry reflects a positive personality trait. COWS-negative: I worrying disrupts effective performance, II worrying exaggerates the problem, III worrying causes emotional discomfort. COWS-positive: IV worry motivates, V worry helps analytic thinking.   Predicting negative repetitive thinking (PTQ): When entering MCQ-pos (Step 1) and all five WW-II subscales as well as the two COWS subscales referring to positive beliefs about worry afterwards (Step 2), the prediction once again is optimized while it can likewise be further improved by using the MCQ-neg as well (Step 3). This time, variance explanation can be improved, by adding the three COWS subscales referring to negative beliefs about worry (Step 4), which causes all factors with positive beliefs to no longer substantially contribute to variance explanation though (see Table 7).
Predicting GAD symptoms (GAD-Q-IV): First, when adding MCQ-pos (Step 1), no significant amount of variance explanation can be detected, whereas adding the WW-II scales in addition to the COWS scales afterwards (Step 2), leads to a significant amount of explained variance, which again can be substantially improved by entering MCQ-neg ( Step 3) and also by including all three COWS subscales referring to negative beliefs about worry (Step 4).

DISCUSSION
The main purpose of this study was to validate German versions of the WW-II and COWS as questionnaire measures of metacognitions related to worry. Our findings suggest that both questionnaires possess good psychometric properties that are comparable to the original English versions. This includes high internal consistency for all subscales as well Table 6 Summary of stepwise multiple regression analyses of PSWQ, PTQ and GAD-Q-IV: β coefficients (SE B).

IVs
Step 1 Step 2 Step 3 Step 4 Step 1 Step 2 Step 3 Step 4 Step 1 Step 2 Step 3 Step 4 as correlational patterns with other questionnaire measures supporting convergent and divergent validity. In addition, the dimensional structure identified for the original questionnaires and predicted by theory, showed good fit with our data for the German versions. For both questionnaires, the proposed five-factor structures showed good fit. All items significantly loaded onto their respective subscale, and each subscale showed good internal consistency. In sum, our findings suggest that WW-II and COWS can be used as reliable and valid measures of metacognitions related to worry.
Although the main aim of our study was the validation of questionnaire measures, we think that the results of the regression analyses are also informative from a theoretical point of view, with regard to the role of metacognitions, specific contributions of negative and positive metacognitions and possible subtypes in the latter case. First, about 50% of the variance in worry, GAD symptoms and repetitive negative thinking could be statistically explained by metacognitive beliefs. This supports the idea put forward by cognitive models of GAD that metacognitive beliefs are crucial in initiating and maintaining worrisome thinking.
Second, the relative contributions of the different metacognition subscales to predicting symptom levels is of interest. Generally, our findings are in line with previous studies' outcomes (i.e.;Thielsch, Andor & Ehring, 2015a;Thielsch, Andor & Ehring, 2015b), revealing negative metacognitions as comparatively more closely linked to worry than positive metacognitions. The same pattern was found when focusing on negative repetitive thinking or GAD symptoms as alternative dependent variables, which underlines the robustness of the finding. Past research using the MCQ-30 has largely focused on the negative metacognitive belief related to uncontrollability of worrisome thinking, showing close relationships between this specific negative metacognitions and worry. Likewise, our study shows a close relationship between this specific MCQ-30-scale and all worry measures. Our findings suggest that additional facets of negative metacognitions may also be important, such as represented in the COWS subscales ''worrying exaggerates the problem'' and ''worrying causes emotional discomfort''. Interestingly, these two subscales explained additional variance over and above the MCQ-30 for the dependent variables RNT and GAD symptoms, but not for the PSWQ. However, as described in the introduction it cannot be ruled out that the strong association between the MCQ-30 and the PSWQ may at least partly be due to criterion contamination as both measures assess uncontrollability of thinking in a similar way (example item of the MCQ-30 scale: ''When I start worrying I cannot stop.''; example item of the PSWQ: ''Once I start worrying, I cannot stop.'').
Third, the MCQ-30 positive metacognitions subscale mainly focuses on beliefs related to worrying as a problem-solving strategy (e.g., ''Worrying helps me to get things sorted out in my mind.'', ''I need to worry in order to remain organized.'', ''Worrying helps me to avoid problems in the future.''). Our findings suggest that additional subtypes of positive metacognitions need to be additionally considered. Among all subscales addressing positive beliefs, the WW-II subscale ''worry protects against negative emotions'' showed the highest and most stable predictive value for all three dependent variables, and as far as the PSWQ is concerned even when controlling for negative metacognitive beliefs. This is in line with theoretical models of GAD suggesting that worry serves to avoid the experience of negative emotions (i.e., Borkovec & Roemer, 1995;Newman et al., 2014). In sum, the relationship between positive metacognitions and worry might be underestimated by using the gold standard MCQ-30 only, as this represents only a restricted operationalization of the concept while not capturing the theoretically postulated subtypes (Dugas & Robichaud, 2012). This consideration is supported by this study's results, showing a substantial improvement of variance explanation due to COWS-positive and WW-II scales over and above the MCQ-30 subscale.
Taken all this together, our findings support the idea that metacognitions related to worry should be assessed using multiple questionnaire measures. In order to provide a more fine-grained analysis of positive metacognitions, the Why Worry Scale -II appears well suited to be given in addition to the MCQ-30. The COWS, on the other hand, has been shown to be a useful measure of negative metacognitions and can be regarded as an interesting addition to the MCQ-30 assessment. As the MCQ-30 is a well-validated measure for metacognitions itself, the questionnaires translated in this study therefore serve as a supplement and as amplification whenever a more fine-grained measurement is needed.
Some limitations are noteworthy when interpreting the results. Most importantly, we evaluated the questionnaires in a non-clinical sample. This is defendable considering the nature of worry as a cognitive phenomenon being conceptualized on a continuum with similar underlying mechanisms in pathological and non-pathological worry (Tallis, Eysenck & Mathews, 1992). Nevertheless, the use of the questionnaires in measuring pathological worry should be further investigated. In addition, it remains to be shown whether results from our online assessment can be replicated using other types of assessment. Reassuringly, however, earlier research has shown that results of web-based assessments are equivalent to traditional assessment format (Naus, Philipp & Samsi, 2009;Skitka & Sargis, 2006).

CONCLUSIONS
Our findings support the idea that metacognitive beliefs are important in the development and/or maintenance of excessive worry and metacognitions should best be assessed using multiple measures assessing different aspects of these constructs. The German versions of the Why Worry Scale-II and the Consequences of Worry Scale can be recommended to be used in addition to the gold standard Metacognitions Questionnaire-30 (MCQ-30). Our results suggest that they are reliable and valid self-report measures for assessing different facets of metacognitive beliefs.