Type of Study

According to the classification performed by ^{Montero and León (2002)}, the present work is considered an instrumental study, as this type of studies aims at the development of tests and apparatus including both their design and adaptation.

Participants

The sample consisted of 910 participants, 543 women (59.6%) and 367 men (40.03%), age range 15-80 years (M= 29.52; DT= 12.25), from the city of Córdoba, Argentina. As regards the participants’educational level, 6.6% were attending postgraduate courses, 19.6% had complete university education, 51.3% had incomplete university education, 12.6% had complete secondary education, 7.4% had incomplete secondary education, and the remaining 3% had complete primary education. Only one participant had incomplete primary education. As regards the current employment status, the sample composition was the following: 40% students (28.80% university students and 11.20% secondary level students), 32.3% employees, 17% professionals (10% fulltime employees and 7% self-employed with no employees), 4% retired or pensioners and informal workers (2% of each category), and the remaining 4% were company owners or partners, full-time employee or self-employed technicians, skilled and unskilled self-employed workers. Only one person was unemployed at the moment of the data collection. The sampling was of accidental type (^{Kumar, 2005}), as the study was carried out in institutions with the permission of their authorities. Concerning instrument administration, all the participants (n= 910) answered the IPIP Five-Factor Domain Scale (^{Goldberg, 1999}). Some participants (n = 229) answered the inventory again a month later to evaluate the punctuation stability (Test-Retest). The NEO-FFI inventory was answered by 209 participants to evaluate the convergent validity. The six scales of the Frequency Inventory on Daily Living Activities were answered by 402 participants to perform the predictive validity study.

Instruments

IPIP Five-Factor Domain Scale (^{Goldberg, 1999}). It consists of 100 items that define the five personality domains: Openness (O), Conscientiousness (C), Extraversion (E), Agreeableness (A), and Neuroticism (N). Each domain is measured by 20 items and is written in phrase form describing typical people’s behavior. A subject is asked to evaluate the degree of precision with which each phrase describes him/her using a scale of five answer options (from “I strongly disagree with this description of myself” to “I agree a lot with this description of myself”). Goldberg reports Cronbach’s alpha values from 0.88 to 0.91 for the IPIP scales and an average correlation between the IPIP scales and the 100 unipolar markers of 0.70 or 0.78 when the correlation for the attenuation was used (IPIP. (s.f.). Restored on 9th March, 2018, from https://tr.im/81f52).

NEO-FFI Personality Inventory (^{Costa & McCrae, 1992}). It consists of 60 items that provide a fast and general measure of the Big Five personality factors. Each scale consists of 12 elements that describe a person’s typical behavior and allows to measure each of the factors. The answer format is Likert type with five options from “total disagreement” to “total agreement” with a certain phrase (Costa & McCrae, 1992). The reliability indices were α = 0.82 for Extraversion, α = 0.70 for Agreeableness, α = 0.80 for Conscientiousness, α = 0.83 for Neuroticism, and α = 0.74 for Openness to Experience.

Frequency Inventory on Daily Living Activities (^{Goldberg, 1999}). Six scales from the Frequency Inventory on Daily Living Activities were used in this study. These scales consist of 54 items that measure six types of recreational activities distributed as follows. Relatively undesirable activities: Drug Use (14 items) and Irresponsibility (seven items); desirable activities: Creativity (11 items) and Friendship (eight items); and relatively neutral activities: Communication (eight items) and Erudition (six items). A participant is asked to mention how frequently he/she has performed certain tasks (e.g., going to a public library) during the past year using a Likert-type scale of five answer options from (1) never in my lifetime, (2) not in the past year, (3) once or twice during the past year, (4) three or more times, but no more than 15 times during the past year, and (5) more than 15 times in the past year. The reliability values of these scales vary from α = 0.67 (Communication) to α = 0.87 (Drug Use).

Procedure

The data collection was performed both collectively and individually. For the collective collection, secondary level students and university students from Universidad Nacional de Córdoba (UNC) completed the instruments in their classrooms with the corresponding authorization from teachers and authorities and the participants’ prior informed consent. For the individual collection, the authors’ friends, relatives, workmates and acquaintances were asked to participate in the study. All of them had previously been told about the aim of the work and the estimated time to complete the instruments, which might vary among one, two, or three weeks. They were also informed about the approximate time to answer each instrument and that they were expected to answer each test with no breaks in time. Then, they were given a closed envelope with the informed consent and the corresponding instruments.

Data Analysis

The Analysis of Missing Values routine from software SPSS version 19.0 was used to evaluate the pattern of missing values (^{Tabachnick & Fidell, 2007}). Then, the sample was divided into two parts at random. The first half (n = 455) was chosen to identify the best markers of the five factors. To select the main 30 items, the 100 items from the IPIP Five-Factor Domain Scale (^{Goldberg, 1999}) were analyzed using exploratory factor analysis (EFA), previously controlling the variance due to Acquiescence (^{Ferrando, LorenzoSeva, & Chico, 2009}). This analysis was performed with the MATLAB program and FACTOR 9.3 (^{Lorenzo-Seva & Ferrando, 2013}). As an external criterion, the correlation between the items of one dimension and the total scores of the same NEO-FFI dimension was considered. As a last criterion, it was tried to represent each scale by different contents. The second half of the sample was reserved to conduct the non-restricted confirmatory factor analysis of the 30 selected items (^{Ferrando & Lorenzo-Seva, 2000}). The variance due to Acquiescence was biased, and the residual correlation matrix was factored through LISREL 8.5. The second half of the sample (n = 455) was reserved to conduct the non-restricted confirmatory factor analysis of the 30 selected items.

Then, an EFA with the total sample (N=910) was performed, from which the item factor saturations and the factor weights were estimated to then determine the factor scores. Moreover, the correlation corrected between the item response and the factor scores was calculated. When doing this calculation, the factor scores of the five factors plus the score in acquiescence was informed. To control the acquiescence, it is necessary a test with reversed items. It should be stated that the interpretation of the scores free of acquiescence is obtained by calculating the factor scores (instead of the scores obtained as the mere sum of item responses). In this way, the content scores are free of acquiescence. On the other hand, it can be interpreted that people with extreme scores in acquiescence have not answered the test seriously.

For the internal consistency study, the factor score reliability was calculated (^{Mislevy & Bock, 1990}) and the testretest method was used to determine the temporal stability. With respect to the convergent validity estimation, the correlations between the factor scores and the direct scores of the five NEO-FFI scales, which are supposed to be similar, were analyzed. Then, an evidence study of concurrent validity was performed comparing the average scores of the scales between men and women, and youth and adults, through a multivariate variance analysis 2 (sex) x 2 (age) (MANOVA). With this analysis, it was intended to compare if the average factor scores of the groups theoretically agree with the personality profiles identified by the literature (^{Cupani et al. 2014}). In accordance with what was proposed in previous studies (^{McCrae et al., 2000}), participants younger than 30 years old were included in the young group. To estimate the effect size, the eta-squared coefficient (η²) was calculated, following as a criterion what was suggested by ^{Cohen (1992)}, that is, that the effect sizes (η²) 1%, 10% and 25% are considered small, medium-size and big, respectively.

Finally, a multiple regression analysis (enter method) was performed with the aim of evaluating the predictive capacity of the five factors on the dependent variable recreational activities, considering sub-samples according to the participants’ sex and age. Depending on the sample size, the statistical power of each model was calculated with the G*Power 3.1 program (^{Faul, Erdfelder, Buchner, & Lang, 2009}).

Missing Case Analysis

The percentage of missing cases was not over 5%; due to this, it was decided to impute the missing data by a central tendency measure (fashion) because this method provides an attractive balance between precision and conceptual simplicity (^{Shrive, Stuart, Quan, & Ghali, 2006}). It was chosen to impute by fashion to try to count on the five (discrete) response options, proper of the scale, to estimate the polychoric correlations.

Development of the 30-item Inventory

Exploratory Factor Analysis (EFA). Firstly, the 100 items were analyzed through a polychoric correlation matrix. The estimation method used was the unweighted least square (ULS), which is a more robust estimator, with less biased estimates, and adequate when samples are not so big (^{Ferrando & Lorenzo-Seva, 2013}). The Káiser-Mayer-Olkin (.883) sampling adequacy method indicated the feasibility of performing the Factor Analysis. Five factors were extracted and the Varimax rotation was used, since, on a theoretical level, the personality dimensions are expected to be independent in the population (i.e., orthogonal). The variance due to AC (Lorenzo-Seva & Ferrando, 2009) had been previously controlled. Because the IPIP 100-item scales are partially balanced, the procedure proposed by Lorenzo-Seva and Ferrando (2009) was used: a) the number of expected factors according to the content and an additional factor related to AC are retained; b) the first centroid is calculated and taken as an estimate of the factor weight of each item on the AC factor; c) this set of estimates (one estimate per item) is used as an target in a congruent rotation to compute the factor loading value of each item content in the AC factor; d) once the factor related to AC is available, the variance explained by this factor is removed, and it was proceeded to identify the factors related to the instrument content on the residual matrix. From this study, it was observed that the factor loading values in the first AC factor varied between -0.03 and -0.71 for Agreeableness, between -0.04 and 0.63 for Extraversion, between 0.08 and 0.62 for Conscientiousness, between 0.14 and -0.84 for Neuroticism, and between 0.00 and -0.50 for Openness. Secondly, a correlation was performed a correlation (Spearman rank) among the 20 items in a dimension (e.g., Neuroticism) and the direct score of the factor theoretically similar to NEO-FFI (e.g., Neuroticism). The item-factor correlations varied between -0.50 and 0.50 for Extraversion, between -0.50 and 0.47 for Agreeableness, between -0.66 and 0.57 for Conscientiousness, between -0.44 and 0.73 for Neuroticism, and between -0.45 and 0.52 for Openness. Thirdly, it was tried to ensure that each dimension was represented by different personality trait contents. As a result of the combination of these criteria (items representing lower AC, higher factor loading values and item-factor correlations, and items representing the different personality traits), the best 30 items were selected. According to these results, it was observed that the IPIP-R-30, once the AC effect was controlled, presents a simple and orthogonal factor structure. Confirmatory factor analysis (CFA). On the basis of the exploratory factor analysis, the 30 items that combined the three strategies mentioned before were selected; they were used for the unrestricted confirmatory factor analysis (^{Ferrando & Lorenzo-Seva, 2000}) on the second sample.

The variance due to AC was biased and the residual correlation matrix was factored using LISREL. To evaluate the model fit the already mentioned criteria were used. On the basis of the saturation factors obtained in the EFA, an item was selected as the marker for each factor. The obtained goodness-of-fit statistics indicated that the model fits adequately to data (GFI 0.96; CFI 0.94; RMSEA 0.05). Then, the total sample (N=910) was used to estimate the item saturation factors, and the factor weights were used to estimate the factor scores. The CFA suggests that the 30 items of the questionnaire present an adequate fit to data. Table 1 shows the factor loadings of the content factors.

Table 1 Rotated factor structure and descriptive indices 

Note: M = Media; SD = Standard Deviation; rc = Corrected Correlation; AC = Acquiescence; EX = Extraversion; NE = Neuroticism; CO = Conscientiousness; AA = Agreeableness; OP = Openness. The saturations over 0.30 in absolute values are presented in bold.

Reliability. The discrimination indices for the 30 items varied between 0.20 and 0.60 (average 0.43), out of which three items presented values below the cut-off point (See Table 3). The reliability coefficients varied between 0.79 and 0.86. For temporal stability, the indices varied between r = 0.71 in the Extroversion factor and r = 0.80 in the Conscientiousness factor (See Table 2). Although the scale reliability was satisfactory as regards both the internal consistency and the temporal stability, it can be seen that, when compared with the 50-item version, the 30-item version registers and average loss of 4.82% in the internal consistency and of 5.75%.in the temporal stability. Validity evidence. In the convergent validity study, the mean correlation was r = 0.61, with a range from r = 0.45 for Agreeableness to r = 0.71 for Conscientiousness (See Table 2). In the group difference study, an effect was observed from small to medium, significant for sex and age. Women presented higher values than men for Agreeableness (η ^² = 0.09), whereas men presented the highest values for Neuroticism (η ^² = 0.01). Regarding age, it was observed that the youth presented higher values than adults for Extraversion (η ^² = 0.01), whereas adults presented the highest values for Conscientiousness (η ^² = 0.01). It can be observed that the five IPIP-R-30 markers evidenced a lower convergent validity with the NEO-FFI scales (5.85%) in contrast with the 50-item version.

Table 2 Reliability indices, convergence studies, and differentiated studies according to sex and age 

Note: N = 910; FCR = Reliabilty with rotated component; T-RT = Test - retest; Conv. = Convergence with NEO-FFI scales. * p ≤ .05; ** p ≤ .001

Table 3 Prediction of recreational activities from the five factors. 

For the test-criterion validity study, in the sub-sample Young Women the personality traits explain between 10% and 18% of the variance (See Table 3). The traits related to Drug Use are Conscientiousness (β= -0.19) and Extraversion (β= 0.24), to Erudition is Openness (β= 0.32), to Communication are Agreeableness (β = 0.18) and Openness (β=0.28), and to Friendship are Extraversion (β = 0.26), Agreeableness (β= 0.19), Conscientiousness (β = 0.19), and Openness (β= 0.19). In the sub-sample Adult Women, the personality traits explain between 13% and 18% of the variance. The traits related to Drug Use are Conscientiousness (β= -0.24) and Extraversion (β= 0.21), to Irresponsibility are Conscientiousness (β= -0.35), to Creativity and Friendship is Openness (β = 0.30 y β = 0.30, respectively). In the sub-sample Young Men, the traits explain between 19% y 29% of the variance. The trait related to Erudition is Openness (β = 0.40), to Communication are Extraversion (β = 0.20) and Openness (β = 0.29), and to Friendship are Extraversion (β = 0.32), Agreeableness (β= 0.25), and Openness (β= 0.25). In the sub-sample Adult Men, the personality traits explain between 15% and 38% of the variance. The traits related to Drug Use are Extraversion (β= 0.27) and Conscientiousness (β = -0.21), to Irresponsibility are Extraversion (β = 0.20) and Conscientiousness (β= -0.36), to Erudition are Extraversion (β = 0.23), Agreeableness (β = 0.24), Neuroticism (β = 0.28), and Openness (β = 0.39), to Creativity are Neuroticism (β = 0.39) and Openness (β = 0.24), and to Friendship are Extraversion (β = 0.30) and Agreeableness (β = 0.38).

Practical Implications

Although there is a slight decrease of certain psychometric properties in the IPIP-R-30 version (lower to 6%) when compared with the 50-item version, the time saving implied in administering a brief scale in certain studies, in which minimizing the evaluation time and the respondent fatigue are vital aspects, compensates this loss. Thus, the time to respond completely to an inventory is estimated between four and ten minutes for a 60-item one, between three and five for a 30-item one, and between two and three minutes for a 15-item one (^{Soto & John, 2017}). This modest time saving can be relevant for certain types of studies. Some examples of studies in which it is really convenient to use brief scales include: the large-scale surveys designed to evaluate different constructs, longitudinal studies requiring that each participant completes the same personality measure in different times, laboratory studies in which part of the time should be destined to experimental manipulations and behavior observations, and pilot studies or research in which factors such as boredom, fatigue, and lack of interest might generate a careless or at random response pattern (^{Credé et al., 2012}; ^{Milojev et al., 2013}; Soto & John, 2017; ^{De Vries, 2013}).

On the other hand, some studies have evidenced that personality tests administered to samples of individuals with low educational level, with lower to average intelligence, adolescents, preadolescents and older adults present the worst fit to the FFM (^{Meisenberg & Williams, 2008}; ^{Rammstedt, Goldberg, & Borg, 2010}; ^{Ross & Mirowsky, 1984}; ^{Soto, John, Gosling, & Potter, 2008}; ^{Vigil-Colet, Lorenzo-Seva, & Morales-Vives, 2015}). In fact, Ramsteed et al. (2010) demonstrated that the FFM factor structure only remains the same throughout the different educational levels when the AC bias is removed. Therefore, the validity of the personality tests in these subpopulations might be compromised due to high levels of AC bias characteristic in these groups, explaining why it is relevant to use scales free of the AC bias, such as the IPIP-R-30, in the evaluation of those individuals.

According to what was previously stated, it was considered adequate to select six items per scale because recent research indicates that the use of a very brief form (e.g., two items) is associated with the increase of Type 1 and Type 2 errors (^{Credé et al., 2012}; ^{Kruyen et al., 2013}). In this way, an instrument was designed with a simple orthogonal structure of five factors, satisfactory reliability indices (internal consistency and temporal stability), evidence of convergent validity with the NEO-FFI scales, group differences regarding sex and age, and predictive validity of recreational activities.

In summary, the results of the present study suggest that both versions represent a valid alternative to measure the personality traits in our population. The results also suggest that the IPIP-R-30 scales have adequate psychometric properties and might be used with relative reliability as an auxiliary tool in research and professional work (clinic, work-related, etc.). Moreover, one of the main advantages of this work is that the obtained scores are free from the AC bias.