Adaptation of the Learned Helplessness Scale in Brazil

Abstract The present study aimed to adapt the Learned Helplessness Scale (LHS) to Brazilian Portuguese and assess its psychometric properties and nomological network in a Brazilian sample. Participants completed an online survey that consisted of the LHS, Rosenberg’s Self-Esteem Scale, and the Perceived Stress Scale. 429 people participated, aged 18-79, mostly women, from multiple regions of Brazil and socioeconomic status. Exploratory and Confirmatory Factor Analysis suggested a unidimensional solution with 18 items as appropriate. Correlations between the LHS and the other instruments also provided evidence of the expected nomological relationships. And we assessed the differences in LHS scores by sociodemographic data. These findings are favorable to the use of this instrument in Brazilian samples.

The concept of learned helplessness (LH) is traditional to psychological studies, dating back to Seligman and Maier's (1967) seminal experiments on the effects of uncontrollable aversive events (i.e., shocks) in dogs.In Portuguese language journals and sites, LH has been cited 247 times according to the Scientific Electronic Library Online (SciELO) Brazil and approximately 1,130 times according to Google Scholar (according to a search done in June of 2021).These results may not even correspond to the totality of Brazilian studies on this subject since publications in English by Brazilian authors were not considered.Although it is a highly known theory, up to this date, we have not found any attempts to create or adapt an instrument to measure LH in Brazil.This study aims to adapt the Learned Helplessness Scale (LHS; Quinless & Nelson, 1988) to Brazilian Portuguese.Such instrument nonexistence in the Brazilian context may be a consequence of two problems with the loss of academic interest in this phenomenon globally.The first concerns the uncontrollable aversive events being insufficient to affect humans' behavior, cognition, and affection, as initially predicted (Miller & Norman, 1979).This was a consequence of the theory's disregard for the role of human cognition and our ability to form, adapt and use schemas to predict and influence future behavior (Miller & Norman, 1979).
Regarding this first problem, the original theory has gone through many revisions, and Abramson et al. (1978) proposed that uncontrollability would lead to LH when a person attributed this to (a) internal factors ("It's me!"), vs. external; (b) global situations ("It'll affect everything I do!"), vs. specific; and (c) stable situations through time ("It'll last forever!"), vs. unstable (Abramson et al., 1978;McKean, 1994).Miller and Norman (1979) also proposed a fourth attribution about the importance of the situation, in which important situations lead to greater LH.However, Abramson et al. (1978) and Quinless and Nelson (1988) considered this attribution represented by the other three.
The second problem concerns LH consequences, which are: behavioral, such as passivity, giving up, and procrastination; cognitive, such as decreased problem-solving ability, frustration, and lowered self-esteem; and affective, such as fear, dysphoria, and depression (Maier & Seligman, 1976;McKean, 1994).As some of its consequences trace back to depression symptoms and previous studies presented a high correlation between LH and depression (see McKean, 1994;Quinless & Nelson, 1988), the theory became known as a depression theory.Although, as pointed out by Costello (1978), it is not a good one, so interest seems to have drifted away from LH to more complex depression theories.
Despite its downfalls in depression research, Fontaine and Faria (1989) argued that LH is a relevant construct, mostly related to the contributions of attribution theory (Abramson et al., 1978).The addition of individual attributional patterns highlighted the dependence of the locus of control and its relevance in generating LH despite a consistent aversive situation, such as within a prison setting (Santos et al., 2007).Faria (1999) also suggests that the development and stability of attributional patterns come from individual and socializing experiences, making them culturally related.Therefore, they should differ based on the individual's context.
In that manner, further research found evidence that LH contributes to greater drug use and substance abuse (Sterling et al., 1996).It was prominent in women, the elderly, and people with low socioeconomic status within a Portuguese sample (Santos & Faria, 2008).It led Native Americans to seek social mobility through identification with majority groups (Schiefer & Krahé, 2014).And it mediated the relationship between social concerns (as an environmental concern) and consistent behaviors (such as donating to environmental causes) (Landry et al., 2018).Those findings point to LH still being relevant in our context, and it seems the research possibilities were not well developed yet.Especially with the COVID-19 pandemic, which affected all aspects of human life.For example, Zheng et al. (2020) found that perceived control (theoretically opposite from LH) led to greater coping with the pandemic effects in a Chinese sample.Also, Šrol et al. (2021) found that perceptions of risk associated with COVID-19, such as cognitive perception of risk and distrust of formal institutions to handle the pandemic, are associated with a stronger feeling of lack of control, and both are also associated with higher endorsement of conspiracy theories.Therefore, we consider there are further possibilities to explore the possible predictive role of LH in different domains of human behavior.Quinless and Nelson (1988) proposed a measure for LH, which accounted for the advances by Abramson et al. (1978) regarding causal attribution effects and kept the uncontrollability dimension.The LHS was supposed to reflect four theoretical factors, although, its dimensionality was never consensual among different studies (see Santos et al., 2007).In Quinless and Nelson's (1988) study, the analysis presented a solution with five factors, one for each dimension of causal attribution and the last two with only two items each.The authors decided to retain those four items because they were related to the concept of LH, although its factors were not interpretable.
Despite that, several authors suggest treating the scale as essentially unidimensional (McKean, 1994;Quinless & Nelson, 1988;Sterling et al., 1996).And in more recent psychometric assessments in a Portuguese context, two studies came to the same conclusion of a unidimensional scale with high reliability (Santos et al., 2002;Santos et al., 2007).
As for its nomological network, aside from the previously commented relations, such as high correlation with depression scales and predictive relations, the LHS is moderate to strongly correlated with self-esteem (Quinless & Nelson, 1988) and perceived stress (Sullivan et al., 2012).Also, the Semantic Scale Network (Rosenbusch et al., 2020) indicated that the LHS maintains somewhat semantical unique items, with the most similar scales related to work (r = .57)and self-worth (r = .51),with the latter also related to self-esteem.
The present study aims to adapt the LHS to Brazilian Portuguese and evaluate its psychometric properties and nomological network in a Brazilian sample.We expect to find a unidimensional solution directly correlated with self-esteem and inversely correlated with perceived stress.

METHOD
This study's pre-registration, data, materials, and syntax are available in the Open Science Framework, https://osf.io/pt45x/.

Translation Procedures of the LHS
We followed the International Test Commission (ITC, 2017) and Borsa et al. (2012) guidelines.Therefore, the scale was initially translated by two independent translators (a specialist in psychometrics and a person of the general public) into Brazilian Portuguese, then these versions were synthesized.The synthesis consisted of combining the redaction of the translated items towards the most adequate contextual and conceptual adaptation.Three psychologists and three people from the general public then evaluated this version in terms of the generalization, clarity, redaction, and understanding of the items (Borsa et al., 2012).After minor changes that did not require a new evaluation, the scale followed for a back-translation performed by a third independent translator.Finally, the back-translated version was sent for appraisal by a psychometrics specialist that deemed it adequate, since the scale authors did not respond in due time.

Participants
Considering the guidelines by Hair et al. (2005), the ratio of variables by observations should be 1:10, and the minimum sample size must be composed of at least 200 participants.However, for the proposed analyses this should be doubled, which increases this study's planned sample size to a total of 400 participants.
Anyone over 18 years old was allowed to participate, although we removed participants that did not complete the survey (n = 869) or failed the attention check items (n = 35).Non-probabilistic (through the researcher's social networks and Facebook Ads) sampling techniques were used, and the analyses consisted of 429 participants, which met the minimum sample size criterion.
Of those, 34.03% were from Federal District, 20.75% were from Sao Paulo, 8.62% were from Rio de Janeiro, and the rest was distributed in other Brazilian states; 80.65% identified as woman and 18.65% as man; aged 18-79 (M = 44.1;SD = 16.4;Mdn = 46); 43.59% were undergraduates, 21.91% were graduates, and 18.41% were high school students, the rest being distributed between postgraduates and middle school.The participants' monthly income was normally distributed and ranged from less than a minimum wage per month (BRL 1,100.00) to more than nine minimum wages per month (> BRL 9,900.00),with a peak of three to five minimum wages per month (BRL 3,300.00-5,500.00).

Instruments
The Brazilian version of the Learned Helplessness Scale (Quinless & Nelson, 1988).The LHS can be used to measure states of learned helplessness and consists of 20 items, answered on a four-point scale (1 = Strongly disagree; 4 = Strongly agree).Original validation study of this scale suggested a structure with five factors (α = .83-.94; Quinless & Nelson, 1988), although its usage and further validation studies suggest an essentially unidimensional structure (α = .90-.97; Santos et al., 2002;Santos et al., 2007).The LHS score is obtained by summing the item scores, ranging from 20 to 80, with scores above 41 indicating learned helplessness (McKean, 1994).

Procedures and Data Analysis
Participants were invited to access an online form available on the EFS platform (https://ww3.unipark.de/www/main.php)through a web link or QR Code.Divulgation was made with an image and text developed by the first author, containing a call to action.The form consisted of: (i) research presentation, information about data treatment, confidentiality, and researchers' identification and contact information, a free and informed consent checkbox was required to proceed; (ii) the SES, a first attention check item, and the PSS-10; (iii) the LHS and a second attention check item; and (iv) sociodemographic data, such as age, gender, education, and monthly income.Participants took approximately eight minutes on average to complete the form.
Attention check items appeared similar to other items in the form.Both items have a statement that read: "This is an attention check item, mark [option] and proceed with the form."The first attention check asked the participant to mark agree, the second attention check, the correct answer was disagree.
In the data analysis stage, the sample was randomly split into two approximate halves to evaluate the LHS factor structure with an Exploratory Factor Analysis (EFA; n = 214) and assess its adequacy with a CFA (n = 215).Given the ordinal nature of the scale, a polychoric matrix was used with 500 bootstrap samplings in both analyses.Also, confidence intervals (CI) are reported.Each sample was assessed for its adequacy with the Kaiser-Meyer-Olkin (KMO) test which is acceptable if greater than .70(Damásio, 2012).
The EFA was implemented through the FACTOR software, version 10.10.03, and the factor retention criterion was the Parallel Analysis with Optimal Implementation technique (PA; Timmerman & Lorenzo-Seva, 2011) and the Hull Method (HM; Lorenzo-Seva et al., 2011) was used to confirm the solution.The Robust Diagonally Weighted Least Squares (RDWLS; Asparouhov & Muthen, 2010) extraction method was used since it is adequate for ordinal variables with normality deviation, and items with loadings under |.30| were excluded.
The H index and the Closeness to Unidimensionality Assessment were also analyzed.The first evaluates how much a set of items represents a common factor, with high H values (> .80)suggesting a well-defined latent variable that is more likely to remain stable in different studies (Ferrando & Lorenzo-Seva, 2018).The second verifies whether the scale can be understood as unidimensional and includes three indices: the Unidimensional Congruence (UniCo), which must be greater than .95; the Explained Common Variance (ECV), which must be greater than .85;and the Mean of Item Residual Absolute Loadings (MIREAL), which must be lower than .30(Ferrando & Lorenzo-Seva, 2018).
Since FACTOR runs a Semi-Confirmatory Factor Analysis (SCFA) along with the EFA, the goodness-of-fit indices derived from that analysis were evaluated as well.For that, we used the Root Mean Square Error of Approximation (RMSEA) which should be less than .08, the Comparative Fit Index (CFI), and the Tucker-Lewis Index (TLI) which should be above .90,or preferably .95(Brown, 2006).
The CFA was implemented using the lavaan package, version 0.6-8, (Rosseel, 2012) for R.Although stated differently in the pre-registration, this analysis was conducted with the RDWLS estimation method as it is more appropriate for ordinal measures and complements the EFA.The same fit indices of the SCFA were assessed.
With the confirmed factor structure for the LHS, we assessed the evidence of nomological validity through the polychoric correlation between the LHS, the SES, and the PSS-10 factor score estimates using the psych package, version 2. 1.3,(Revelle,n.d.) for R. Also, we assessed exploratory differences in LHS scores by sociodemographic data through t-tests and robust ANOVAs with the WRS2 package, version 1.1, (Mair & Wilcox, 2020) for R. Considering the robust ANOVAs computed by this package, the procedures consider: (a) for the test statistics, the trimmed means and allows heteroscedasticity; (b) for the explanatory effect size measure (ξ), it does not require equal variances and can be generalized to multiple groups settings, its interpretation consider .10corresponding to a small effect size, .30 to a medium effect size, and .50 to a large effect size; and, (c) for the posthoc tests, it is computed on trimmed means and use the linear contrast expression, represented by Ψ̂ (Mair & Wilcox, 2020).

RESULTS
The polychoric correlation matrix was considered appropriate for the EFA with a KMO of .86,95% BC CI (Bias Corrected CI) [.825, .826].The PA technique suggested a unidimensional solution, considering the 95 percentile of random variance, and a solution with two factors considering the mean of random variance, therefore we used the HM which confirmed the unidimensional solution.
The first analysis was conducted with 20 items which presented an explained variance of 47.44% and good reliability indices, although two items had poor loadings.Item 4 ("I don't place myself in situations in which I cannot predict the outcome"; loading of .01)and 19 ("I feel that my success reflects my ability, not chance"; loading of -.23) had loadings smaller than the cutoff value (< .30).Thus, they were excluded from the scale, and a new analysis was run.Table 1 presents the item loadings, explained variance, eigenvalue, and reliability coefficients for the analysis with 18 items.
Considering the 18 items solution, the second half of the sample was assessed for its adequacy.The polychoric correlation matrix was deemed appropriate, KMO = .84,95% BC CI [.83, .86],and the CFA was conducted.Goodnessof-fit indices for both the SCFA and the CFA were fair, with slightly better results for the first, as it is a less restrictive analysis (see Table 2).Also, all reliability estimates for the CFA solution were good, α = .94,ω = .96and CR = .97.
Then, we assessed the convergent and divergent validity using the total sample (N = 429) between the factor score estimates composed of a polychoric correlation.As expected, the LHS was strongly and positively correlated with the SES, r = .86[.83, .89],and the LHS was negatively correlated with the PSS-10, r = -.72 [-.77, -.67].
For the difference in the LHS scores by sociodemographic data, we also considered the total sample for the analysis and computed the scores by summing the item scores, as proposed by McKean (1994).
Considering gender, the t-test was computed for people who identified as woman and man, non-binaries and other categories were discarded due to the small group size (n = 3).On average people who identified as man presented slightly higher levels of LH (M = 50.88,SD = 8.28) than people who identified as woman (M = 49.40,SD = 8.04), although those results were non-significant, t (NA) = -1.006,p = .31,95% BCI [-2.97, 1.07].
As for the state the participants were from, the categories were computed by region to assure larger groups and representation, only the Central-West (n = 181) and Southeast (n = 158) regions of Brazil had relevant group sizes, and other respondents were filtered out of the dataset (n = 90).For these groups, results were non-significant, t (NA) = .48,p = .62,95% BC CI [-1.35, 2.25], as the mean difference between groups was close to zero (.45).

DISCUSSION
This study aimed to adapt the LHS to Brazilian Portuguese and assess its psychometric properties in a Brazilian sample.The multiple procedures assessed in the EFA (i.e., PA, HM, and Closeness to Unidimensionality Assessment) indicated that the unidimensional solution is adequate for a Brazilian sample.This finding reproduces other studies that suggest the usage of the LHS as a general LH measure (McKean, 1994;Santos et al., 2002;Santos et al., 2007;Sterling et al., 1996).
In the original 20-item version of the scale, items 4 and 19 presented factor loadings under the cutoff value (.01 and -.23), so we removed them.Item 4 already displayed poor factor loading in the Portuguese sample (.06 in Santos et al., 2007) and could be caused by double negatives in the affirmative, which demand more cognition from the participants and tend to generate inconsistent answers1 .As for item 19, the verb change from "reflects" to "depends" in the Portuguese version could have changed the sentence relation between "success" and "ability" from consequence to dependence.In that manner, participants who scored higher or lower in this item could both indicate greater LH.
The goodness-of-fit indices from the SCFA and the CFA indicated the adequacy of the scale with 18 items following Brown's (2006) cutoff values.The H-index suggested that the latent variable is well-defined and will probably be stable across multiple studies and samples (Ferrando & Lorenzo-Seva, 2018).The scale also showed good reliability through all the assessed indices (i.e., Cronbach's alpha, McDonald's omega, and CR).Evidence of nomological validity was found, as predicted and showed in previous studies, with LHS being strong and positively correlated with SES (Quinless & Nelson, 1988) and strong and negatively correlated with PSS-10 ( Sullivan et al., 2012).
Nonetheless, the factor score of the original 20-items scale was suggested to be the sum of each item's score, which should range from 20 to 80, with scores higher than 41 indicating LH (McKean, 1994;Quinless & Nelson, 1988).Following the same logic and considering that two items were removed from the Brazilian version of the LHS, the factor score should range from 18 to 72, with scores higher than 37 indicating LH.Despite the sum of item scores being adequate for its simplicity, preservation of raw data variation, and retention of scale metrics (DiStefano et al., 2009), its use in classifying or diagnosing individuals as helpless, or not, with only a mean plus one cutoff is not adequate.A better solution for those cases would be to use a standardization sample with at least three categories (e.g., low, medium, and high), but that exceeds the scope of this study.
Furthermore, we did an exploratory assessment of the differences in LHS scores by sociodemographic data.We found no significant differences between people who identified as man and woman.This is consistent with

Table 1
Exploratory Factor Analysis of the LHS with 18 items.

Table 2
Goodness-of-fit Indices of the 18-item Version of the LHS.Note.χ 2 = Robust Mean and Variance-Adjusted Chi Square; df = Degrees of Freedom; BC = Bias Corrected.* p < .001.