Participants

The sample was selected on a non-probabilistic basis and was formed under the following inclusion criteria: a) having experienced at some point in their lives the effects of a natural disaster (earthquake and/or flood), b) being of legal age, c) having freely approved their participation, d) being a member and resident of these communities. The post-data collection exclusion criteria were unanswered items (missing values) and the identification of multivariate extreme values (outliers).

The selected sample, based on non-probabilistic sampling, consisted of inhabitants of Jojutla (n= 263, 91 %) and Yautepec (n = 27, 9 %), with a total of 290 participants, 55.7 % women and 44.3 % men with an age range of 18 to 79 years (M = 29.4, SD = 14.5). Regarding educational level, 27.9 % had basic education, 39.7 % had high school, 30.2 % had a bachelor's degree, 0.7 % had a graduate degree, and 1.4 % had no studies.

Instrument

In the elaboration of the situations that promote community resilience we relied on the theoretical components proposed by ^{Twigg (2007}) and ^{Suárez-Ojeda (2001}), considering the eight dimensions explained above.

Procedure

Dimensionality. The instrument was represented by eight dimensions based on the 5 theoretical dimensions proposed by ^{Twigg (2007}) and the 3 proposed by ^{Suárez-Ojeda (2001}), with correlated expectations. The definitions were elaborated considering their conceptual applicability to the context of the chosen Mexican communities and their potential for generalization to other Latin American contexts, their approximate convergence with other definitions in the literature, their potential to be represented in the content of the items (^{Delgado-Rico et al., 2012}).

Item development. To write the items, the existing literature was reviewed, determining the theoretical model and the operational definition of the dimensions that would make up the instrument. Subsequently, the items were written in the first person, taking care to use clear, homogeneous language in accordance with the proposed dimensions.

As a guideline for response options, a 5-point Likert scale was proposed to give as possible responses the degree of disagreement-agreement, in relation to the items elaborated. A homogeneous length was sought in terms of content and appearance, preventing items from overlapping each other (^{Martínez et al., 2009}). Initially, 27 items were proposed, distributed in 8 theoretical dimensions as shown in table 2.

Table 2: Conceptual operationalization and distribution of items of the instrument 

Expert judges. Once the instrument proposal was integrated, the expert judgment technique was used to submit the questions to the scrutiny of experts in the field, to determine how relevant, the questions of the scale were in relation to the proposed topic (^{Escobar-Pérez & Cuervo-Martínez, 2008}). The group of judges consisted of five psychologists chosen for their verifiable expertise through their publications and field experience in their areas of work in relation to the topic: two full-time academics at universities in Mexico and Chile, and three disaster psychologists who have been part of the Mexican and International Red Cross and the Psychological Attention for Disasters (APSIDE) program. Each was contacted individually and sent a form explaining the instructions for evaluating the content of the items on the criteria of coherence (defined as that the item has a logical relationship with the dimension or indicator being measured), relevance (defined as that the item is essential or important, so it should be included) and clarity (defined as that the item is easily understood, i.e., it has adequate syntax and semantics). They were also asked to make observations and suggestions according to their consideration. Each judge made his or her contributions independently of the other judges. The judges' responses were analyzed by means of ^{Aiken's V coefficient (Aiken, 1985}) to see the degree of agreement among the experts in terms of clarity, relevance and coherence. The ICAiken program (^{Merino-Soto & Livia, 2009}) was used.

Participating judges. Because there may be possible discrepancies in the perception of the conceptual relevance and clarity of the items perceived by judges and participants (^{Merino-Soto, 2016}; 2018). A pilot application was carried out to 33 people from one of the communities to be studied, with an age range of 18 to 70 years (M = 34.94; SD = 17.97), of which 60.6 % were men and 39.4 % were women. During the application, the participating judges made observations on the instrument, regarding words that for them had a different meaning or that were not clear and caused confusion in their answers.

Regarding the proposed dimensions, they were unable to differentiate between the dimensions of community support and coping; they considered that they belonged to the same dimension. The humor dimension was not clear to them, and they did not consider it important.

Data cleaning. Data cleaning began with the identification of a) the number of unanswered items in the instrument, and b) multivariate extreme values. The first was done with a simple count of the unanswered items; the second was done by applying the D2 statistic (^{Mahalanobis, 1936}), which obtains the multivariate distance of the data from the centroid response of the variables. The program norm (^{DeCarlo, 1997}) was used. For descriptive statistics, and univariate (Cramér-von-Mises test, CVM; ^{Darling, 1957}) and multivariate normality (HZ test; ^{Henze & Zirkler, 1990}), the MVN program (^{Korkmaz, et al., 2014}) was used.

Dimensionality. The analysis of the internal structure of the instrument consisted of a combination of exploratory and confirmatory approaches to establish the structure of the instrument. The exploratory approach was applied to verify the number of latent dimensions underlying the data, and to identify the existence of possible redundant dimensions. The number of dimensions proposed for the instrument was six, which served as a hypothesis of dimensionality. A consensus approach (^{Lüdecke et al., 2020}) was used among several dimensionality identification methods, using the programs nFactors (^{Raiche & Magis, 2020}) and psych (^{Revelle, 2019}), integrated into the parameters program (^{Lüdecke et al., 2020}). This approach consists in the maximum convergence of 24 methods regarding the number of possible dimensions.

Internal structure of the items. In the next step, a semi-confirmatory factor analysis was applied (^{McDonald, 2005}). Although in the exploratory stage unidimensionality was suggested as the recommended internal structure, the semi-confirmatory factor analysis was applied to observe the congruence of the items with their factors and the degree of factorial complexity. The Factor 10.10 program (^{Lorenzo-Seva & Ferrando, 2006}) was used, applied to the polychoric correlations between items. To evaluate the congruence between the items with their expected relationship to the factors, the congruence coefficient (^{Burt, 1948}) was used; values greater than 0.90 indicate good congruence (^{Lorenzo-Seva & Ten Berge, 2006}), so we proceeded to review the items that met the congruence index, accepting 16 of these and although unidimensionality was suggested, in the final scale we considered that at least two items from each of the theoretical dimensions should be included.

Data analysis

Expert judges. All expert judges invited to evaluate the items completed the task. Of the 27 items evaluated, only 1 item was considered to have insufficient content validity (clarity, coherence and relevance < .87) and was therefore eliminated. Improvements were made to the wording of 3 items and 1 item was added. Table 3 shows the results obtained with the Aiken V for clarity, coherence and relevance.

Table 3: Degree of agreement among the judges in relation to clarity, coherence and relevance 

^Note: V: AIKEN V Validity Coefficient.

Participating judges. All the participating judges completed the task. They reported that they did not associate the word "disappeared" with the topic of natural disasters but related it to situations of violence and organized crime. Likewise, the questions on the dimensions of humor, community coping, community support and solidarity were not considered relevant, caused them confusion and they mentioned that they were similar.

Subsequently, with the judges' observations and suggestions, it was decided to merge the dimensions of community coping and community support and solidarity into a single theoretical dimension since the wording and content of these dimensions were very similar and their separation only confused the reader. Then, the questions related to the use of humor as a way of coping with danger were eliminated as they were evaluated as irrelevant. At the judges' suggestion, four more items were added to the knowledge and response readiness dimensions, so that the scale for validation consisted of 24 items.

Preliminary analysis. Three cases (1 %) with no response to the items were detected, 2 (6 %) were imputed by the modal value of the item. Regarding the multivariate extreme values, by means of the Mahalanobis distance, 10 participants were detected and eliminated, representing 3.3 % of the sample.

Descriptive analysis of items. The descriptive results of the items are presented. The set of items did not show multivariate normality (Henze-Zirkler test = 1.211, p < .01); this was derived from the absence of univariate normality in each item (Cramer-von Mises test, CVM < .01). On the other hand, there seems to be no restriction in the response range, because the participants used the entire range of options. This similarity in the use of scaling was also observed in the dispersion (standard deviation) of each item as shown in table 4.

Table 4: Descriptive statistics for the Community Resilience Scale items (n = 290) 

^Notes: ERC: community resilience scale; CVM: Cramer von Mises univariate normality; As: skewness coefficient; Cu: coefficient of kurtosis. *p < .05.

Dimensionality. The empirical eigenvalues obtained from the inter-item correlation matrix were as follows (top seven): 10.379, 1.627, 1.105, 1.031, .988, .906, .797. With the application of the consensus approach it was found that a single dimension was suggested by 21.7% of methods based on the hillside plot or scree test (e.g., acceleration factor and multiple regression), VSS (method 2), and model fitting (e.g., TLI and RMSEA coefficients); two factors (8.6 %) by VSS (method 1) and MAP-Velicer; three factors (13.0 %) by CNG, BIC-information and BIC-fitting; 4 factors (21.7 %) by methods predominantly based on the slope graph. Due to the parsimony in the construct representation of the instrument, the difference between the first eigenvalue compared to the rest of the eigenvalues, and that the eigenvalues from the second were comparatively very small with respect to the first, the apparent number of factors suggested for the measurement model was a single dimension.

To examine and verify the apparent strong latent unidimensionality of the instrument items, the factor loadings of the 24 items are shown in table 5. There are items that load on more than one factor, and the correlations between dimensions are considered high, so it was decided to perform a semi-confirmatory analysis.

Table 5: Factor loadings per ítem 

Table 6 shows the level of congruence with the unidimensionality and the factor loadings, where those that had an I-Unique and I-ECV = or < .95 were considered to be part of the final instrument, forming a unidimensional scale with 16 items that will allow the general measurement of Community Resilience in populations of the state of Morelos that have been exposed to earthquakes and floods.

Table 6: Level of congruence of unidimensionality 

Also, the 16 items that made up the final instrument, with factor loadings between .478 and .850 (M= .693; Md = .711) are shown in bold. The reliability coefficient Ω was .924. Therefore, it is considered to meet very high values of reliability and internal structure validity.