Participants were included in four national programs promoted in Colombia by the “Prosperidad Social”, Department of Colombian National Government: Ingreso Social (participants from 18 to 35 years of age), Jóvenes en Acción (participants from 16 to 24 years of age), Enrútate TU (participants from 30 to 60 years of age; during the application of the assessment tools, this program was converted into Ruta de Inclusión Productiva) and Habilidades para el futuro (participants from 18 to 55 years of age). Beneficiaries of these programs are people in situations of extreme poverty or victims of forced displacement, due to armed conflict or violence related to drug trafficking. Participants in the first three programs lived in eleven different town halls, in order to ensure adequate representation: Florencia, Neiva, Barranquilla, Manizales, Bucaramanga, Medellín, Sincelejo, Cali, Buenaventura, Montería, and Cartagena. Participants in the last program (Habilidades para el futuro) lived in the following three town halls: Pasto, Popayán and Riohacha.

Colombian participants were 947 young and adults (427 males, 520 females), ranging in age from 16 to 66 years (M = 32.4, SD = 11.6). Their socioeconomic backgrounds varied widely, depending on their geographical position: 76% were residents of an urban area, and the remaining 23.9% lived in rural areas. Overall, 21.6% completed junior high school and 57.8% completed high school. Only 39.1% had a job at the time of this study. Overall, 90% of participants had been victims of forced displacement because of armed conflict or violence related to drug trafficking.

The present investigation was based on data collected as part of a larger study, including vulnerable individuals taking part in the above-mentioned projects. Data were collected with paper-and-pencil self-report questionnaires. The completion of the questionnaire required approximately forty-five minutes to one hour in total. All adult participants signed a written informed consent. Written informed consent was provided by parents of underage participants. A team of data collectors, composed of three trainees in psychology, was present throughout the administration of questionnaires, which took place in the headquarters of the various programs.

The Department of Education of Roma Tre University does not have an ethics committee. However, this study was conducted in accordance with the requirements of privacy and informed consent laid down by Italian law (Law Decree DL-196/2003). Moreover, the study adhered to the latest version of the Declaration of Helsinki revised in Fortaleza (World Medical Association [WMA], 2013) and followed the ethical guidelines of the Italian Association of Psychology (AIP). An Italian researcher and a Colombian researcher were available during administration in all programs, in order to assure comprehension of the items.

Ego-resiliency. The ER89-R (Vecchione et al., 2010; Raciti et al., 2015) is a brief inventory composed of 10 items. The psychometric properties of this instrument are presented in Table 1. Cronbach’s Alpha was 0.85 for the total scale and 0.78 and 0.75 for Optimal Regulation and Openness to Life Experiences respectively.

Self-esteem. It was assessed with 5 items from the Rosenberg Self-Esteem Scale (Rosenberg, 1965). In general, this scale assesses the extent to which participants believe they possess good qualities, accept their own characteristics, and have achieved personal success or experienced failure. For each item, ratings are provided on a 4-point scale ranging from 1 (strongly disagree) to 4 (strongly agree). A sample item is, “I feel that I have a number of good qualities.” Internal consistency was good with Cronbach's alpha = 0.87.

Optimism. It was assessed with the Life Orientation Test (Scheier, Carver, & Bridges, 1994). We used only six items of 10 total scale (four items were fillers) that measure respondents’ expectations about the future. Participants provided their ratings on a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree). A sample item is “In uncertain times, I usually expect the best.” Internal consistency was satisfying with Cronbach's alpha = 0.70.

Coping. It was assessed with the COPE questionnaire (Carver, Scheier & Weintraub, 1989) measuring coping strategies that people use when they face a negative and stressful event. The questionnaire consists of three subscales: Problem-Focussed Coping (to strengthen their efforts to eliminate or reduce stress); Emotion-Focussed Coping (to use emotional support based on sympathy and understanding); Disengagement (to use alcohol or other substances with the aim to feel good or to endure stress). We centred our attention on Problem-Focussed Coping with its subscales: Active coping (α = 0.50), Planning (α = 0.73), and Use of instrumental support (α = 0.60). The complete questionnaire consists of 45 items rated on a 4-point scale: 1 = never; 2 = rarely; 3 = sometimes; 4 = always.

In order to assess the psychometric properties of the scale in a Columbian population, we conducted a confirmatory factor analysis using Mplus 8.1 (Muthén & Muthén, 1998-2018), using maximum likelihood (ML) minimisation functions and then, after having established a well-fitting structure, multi-group CFAs were used to test the hypotheses regarding measurement invariance across gender and across age.

We relied on common fit indices to evaluate model to data fit: χ², comparative fit index (CFI), Tucker-Lewis Index (TLI), root mean square error of approximation (RMSEA), and standardised root mean squared residual (SRMR). Using a chi-square to test for differences among models is frequently used. However, this test is too sensitive for a large sample size (N = 300) (Kline, 2011; MacCallum, Browne, & Cai, 2006). Owing to this reason, we relied on the principle that the χ²/df ratio should be smaller than 3 (Schermelleh-Engel, Moosbrugger, & Müller, 2003) and made use of the alternative criteria (Akaike Information Criterion [AIC]; see Brown, 2015) to compare models. CFI and TLI values greater than .90 (Kline, 1998), RMSEA values lower than .07 (Browne & Cudeck, 1993), and SRMR values lower than .08 (Kline, 1998) are considered adequate for good models. Lower values of AIC indicate a better model fit than compared models (Brown, 2015).

In order to test the measurement invariance, a model-fitting process was adopted based on the review by Vandenberg and Lance (2000). Five consecutively more restrictive nested models were tested (Vandenberg, 2002): configural invariance (the same pattern of fixed and free factor loadings across groups), metric invariance (equal factor loadings for items across groups), scalar invariance (the same factor loadings and latent intercepts for the items across group), and strict invariance (imposing an additional constraint on item residual variances). At this regard, we followed Chen’s (2007) simulation study, which suggests that ∆CFI, ∆RMSEA, and ∆SRMR could be used to inspect changes in model fit between nested models. The difference of fit indices between these models should be smaller than .010 for ∆CFI, .015 for ∆RMSEA, and .030 for ∆SRMR in a large sample size (N = 300) (Chen, 2007). Given that full MI models could not be found, we tested for partial MI. We followed Schwartz and colleagues’ (2014) suggestion that partial invariance is still tenable if less than half of item are freely estimated.

Finally, we tested the construct and discriminant validity of the ER89-R in a Columbian population. At this regard, we examined the relations of the ER89-R with measures of adjustment (self-esteem, optimism, and three coping strategies). In particular, we conducted a series of standard multiple regression analyses including Ego-resiliency as predictors and one measure of adjustment as the dependent variable (in total, we examined five regression analyses).

We examined the data for univariate and multivariate variable distributions using the procedure devised by Tabachnick and Fidell (1989). No outlier participants were detected. Table 1 presents the means (M), standard deviations (SD), skewness, and kurtosis parameters for the ER89-R items. The Keiser Meyer Olking test of sampling adequacy was 0.90. Bartlett’s Test of Sphericity was also largely significant.

Table 1 Table 1 Descriptive statistics of Items of ER89-R Questionnaire

To investigate the stability of the ER89-R factor structure, we ran separate CFAs using the sample covariance matrices. We tested three models: a one-factor model (Model 1), a two-correlated-factor model (Model 2) and a model with a second order factor (Model 3). The fit indices related to the aforementioned models are reported in Table 2.

Table 2 Table 2 Fit Indices for Confirmatory Factor Analysis of the ER89-R scale Note.

Model 1 refers to one factor model; Model 2 refers to two-correlated-factor model; Model 3 refers to a model with a second order factor; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = Root-Mean-Square Error of Approximation; CI = Confidence Interval; SRMR = Standardised Root-Mean-Square Residual; AIC = Akaike Information Criterion.

Results showed the presence of a second order factor, because this model fit the data better than the other two models, χ² (df = 32; N = 947) = 182.371, p < 0.01, CFI = 0.95, TLI = .93, RMSEA = 0.07 (CI = 0.06 - 0.08), SRMR = 0.04, confirming Block’s theory (Block & Block, 1980; Block & Kremen, 1996). All parameters were significant, p < 0.05, and all were > 0.30. Two modification indexes (MI) were reported for the covariation between error items 8 and 7 (MI = 47.23), and 8 and 9 (MI = 45.64). For identification purposes, we constrained as equal the error variances for the first-order factors (ψ1.1 = ψ2.2), fixed the first loading on the second-order factor (i.e., γ1.1 = 1), and estimated the second-order model that presents an identical fit and degrees of freedom as the previous two-factor model (Bollen, 1989; see Figure 1).

Figure 1

Note. e1-e10 represent error terms. Standardised factor loadings are shown on the straight arrows; error term intercorrelations are shown on the curved arrows. The intercorrelations of the error terms between items 7 and 8 and between 8 and 9 were freed.

Table 3 shows results of MI analyses across genders for the best-fitting model (a second order factor model). Findings confirmed full configural and metric invariance, indicating that the best-fitting factorial structure was confirmed within males and females. Instead, the full scalar invariance was not supported. At this regard, we found that if the equality constraint was lifted from one item intercept for optimal regulation dimension (Items 7) and one item intercept for openness to life experiences (Item 5), then the partial scalar invariance could be established.

Finally, strict invariance and partial strict invariance were investigated. Analysis showed that if the equality constraint was lifted from one residual variance for optimal regulation dimension (Items 10) and one residual variance for openness to life experiences (Item 3), the partial strict invariance could be supported.

Table 4 shows results of MI analyses across three age group (16-25 years, 26-40 years and over 40 years) for the best-fitting model (a second order factor model). Findings confirmed full configural invariance, indicating that the best-fitting factorial structure was confirmed within three age group. The full metric invariance was not supported. At this regard, also the partial metric invariance was not tenable because less than half of intercepts were freely estimated (Schwartz et al., 2014).

Table 3 Table 3 Fit Indices for Measurement Invariance across gender Note.

df = degrees of freedom; CFI = comparative fit index; TLI = Tucker-Lewis Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square residual; AIC = Akaike Information Criterion; ∆χ² = chi-square difference test.

Table 4 Table 4 Fit Indices for Measurement Invariance across age group Note.

df = degrees of freedom; CFI = comparative fit index; TLI = Tucker-Lewis Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square residual; AIC = Akaike Information Criterion; ∆χ² = chi-square difference test.

Table 5 contains the zero-order correlations between Ego-resiliency and some relevant indicators of adjustment (self-esteem, optimism, and coping strategies). All variables were highly and positively correlated to each other. The correlations were similar for ER89–R Optimal Regulation and for ER89–R Openness to Life Experiences.

The results of the standard multiple regression (see Table 6) demonstrated that the total (second order) score of ego-resiliency explained between 1% and 52% of variance in all adjustment outcomes. As expected, this score predicts the outcomes sufficiently well.

Table 5 Table 5 Zero-order Correlations between Ego-resiliency and Relevant Indicators of Adjustment Note.

**p < 0.01

Table 6 Table 6 Results of Standard Multiple Regression of the Ego-resiliency on Relevant Indicators of Adjustment Note.

**p < 0.01.