Revista Colombiana de Psicología

Participants

Participants were 292 Colombian undergraduate psychology students from a private university in Bogotá (210 female, 71.9%). Age ranged from 18 to 34 years (M = 20.3). Sample size is consistent with guidelines recommending an approximate sample size of N = 300 for a factor analysis (Field, 2005; Tabachnick & Fidell, 2012).

Measures

Procedure

Ethical consent was obtained from the ethics committee in the university at which this study was conducted. Following established guidelines (Sousa & Rojjanasrirat, 2010), a translation of the PSI-Q from English to Spanish was prepared by two bilingual and bicultural individuals whose mother tongue is Spanish (the second author and a postgraduate student). The two versions were compared by another student and discrepancies were resolved as a team. Back translation was then carried out by the first author, whose first language is English, and ambiguities were revised and resolved by the team. At this stage, items were not adapted or added because (i) we wanted to explore the properties of the original PSI-Q, and (ii) all items were deemed familiar and comprehensible to a private-university student population, who have been exposed to international cuisine, activities, etc. Recruitment took place in psychology lectures. All data was collected via pen-and-paper. In the initial session, all participants completed a consent form and the PSI-Q (short version), and a subgroup of N = 160 completed the VVIQ for convergent validity. Participants were given a snack to thank them for participating.

Six weeks later, the PSI-Q (short version) was reapplied to 65 participants (two classes, selected randomly) to measure test-retest reliability. Age (M = 19.94, SD = 2.86) and gender (49 females, 75.4%) were similar to that of the original sample. The data is available through open-access repository [ https://osf.io/qrb6j/].

Data analysis

Analysis was conducted using RStudio version 1.1456 (RStudio Team, 2016) with the Lavaan library (Rosseel, 2012). First, we ran a confirmatory factor analysis and checked goodness-of-fit indices and factor loadings to examine whether the factor structure replicated that of the English version. We then proceeded to adjust this model to improve the overall fit. Notice that for the confirmatory factor analysis we have used one of the robust maximum likelihood estimators (MLF) which allowed us to run the analysis with incomplete data.

Internal consistency

Internal consistency was excellent for the PSI-Q total (α = .90, ω = .91) and acceptable to good for Visual (α = .78, ω = .79), Auditory (α =. 60, ω = .62), Olfactory (α = .75, ω = .77), Taste, (α = .73, ω = .75), Tactile (α = .71, ω = .72), Bodily Sensations (α=.68, ω = .68), and Emotion scales (α = .70, ω = .71). Mean correlation between the scales was of 0.53, ranging from 0.32 (p < 0.01) for the correlation between Taste and Vision to 0.6 (p < 0.01), for the correlation between Taste and Smell.

Convergent validity with the VVIQ

The VVIQ scale exhibited a good reliability (α =. 85, ω = .88). Moderate convergent validity was seen between the VVIQ and the original PSI-Q full scale (r = 0.53). All PSI-Q subscale scores were positively associated with the VVIQ, with the Visual and Bodily Sensations subscales having the highest and lowest correlations (r = .67 and r = .31), respectively.

Test re-test reliability

Eight-week test-retest reliability, measured with 65 participants, indicated a significant correlation (r = .65), with no significant differences between the means of both applications (Mpre = 256.78, SD = 26.3, Mpost = 251.98, SD = 26.3), paired t-test: t(64) = 0.73, p = .46, M diff = 4.8, 95% CI [-8.4, 18.01]).

Confirmatory factor analysis of short version

A confirmatory factor analysis offered adequate support for the original model of seven distinct scales. Although the model fit is acceptable for a replication, it was further improved by removing two items with potentially high cross-loadings, “burnt wood” (Smell) and “walking briskly in the cold” (Bodily Sensation), and allowing for residual covariances that include the Smell scale items (See Table 1 for a summary). This may reflect that experiences of wood-burning and cold weather are uncommon in Colombia. The residual covariances with Smell scale items stemmed from high cross-loadings between Smell and other subscales, particularly Taste, resembling findings of the original PSI-Q validation study (Andrade et al., 2014).

The best-fitting model performs better in every single indicator assessed, with better standard fit indices, like the CFI (0.94) and TLI (0.92), and a good RMSEA (0.050). Full factor loadings for the best model are presented in Electronic Supplementary Material 1.

Participants

Ethical consent for the second study was obtained from the institutional IRB at Universidad de los Andes (ref #14638). Participants in the second study comprise low-income adults based in Bogotá who had signed up for an entrepreneurship training program that incorporated aspects of mental imagery. All participants completed a consent form. The data is available through open-access repository [https://osf.io/qrb6j/].

The present study used data collected at baseline on two occasions. First, in April 2019, 43 participants (22 male, 21 female) between 19 and 62 years old (M = 40.56, SD = 13.68) were recruited for a pilot phase of the entrepreneurship training program. Second, in June 2019, 508 adults (261 women, 259 men) were recruited for the main phase of the same project. These participants represent a relatively varied sample of the urban Colombian population. The majority (62.3%) were aged 18-28 years, with 31.8% aged 29-59 and 5.9% aged 60+. In terms of education level, 187 (36.8%) had at least an undergraduate university degree, 172 (33.8%) had a technical (non-university) qualification, 91 (18.1%) had completed high-school, and 58 (11.4%) had not completed high-school. It is worth mentioning that the entrepreneurship program prioritized recruitment of vulnerable individuals, such as victims of the Colombian armed conflict, migrants, and low-income households. However, PSI-Q subscale means in the present study were very similar to those found in the original PSI-Q validation (Andrade et al., 2014), suggesting that participants did not suppress imagery when completing the PSI-Q.

Measures

Procedure

The team who adapted the PSI-Q consisted of bilingual Colombian and British researchers living in Colombia, experienced in implementing interventions with participants from across the spectrum of socio-economic status.

In the pilot, participants completed the validated Spanish version of the PSI-Q, as in Study 1. Items were administered verbally via one-on-one interviews by trained research assistants to enable evaluation of participants’ comprehension item-by-item. Research assistants read instructions and items aloud, and visually presented the response scale each time (see final version at the end of this article). The research assistants identified any items that had been difficult for participants to understand.

A new version of the PSI-Q adapted for a Hispanic American context was then developed based upon the insights of the pilot (see results) and applied in the second phase. Data collection used the same approach as the pilot (one-on-one verbal administration) to ensure that low literacy did not affect validity. Researchers supervising data collection ran programmed quality checks daily to correct for duplicates and identify any deviations from the protocols. They were also in regular contact with field staff to clarify possible misunderstandings and conducted random spot checks to ensure that surveys were being conducted according to protocols.

Data analysis

As in Study 1, analysis was conducted using RStudio version 1.1456 (RStudio Team, 2016) and confirmatory factor analyses ran with the Lavaan library (Rosseel, 2012) with an ML estimator. We fitted several models to contrast the original, shortened, and modified versions of the PSI-Q, and explored some moderators of imagery vividness.

Results from pilot study and subsequent adaptation

Mean vividness scores indicated satisfactory ability to imagine (≥ 7 of 10) for most items, except for a “stuffy room” (Smell) with mean vividness of 6.6. As in Study 1, items showed a positive asymmetrical distribution that suggests the task was relatively easy for the participants.

Through group discussions and analysis of pilot data, several of the PSI-Q items were hypothesized to have low face validity for a representative sample of Hispanic American population. Specifically, these were: “mustard” and “black pepper” (Taste), since these condiments are not commonly used in Hispanic American cuisine; “relaxing in a warm bath” (Bodily Sensation), since baths are rarely found outside upper-class homes; and “walking briskly in the cold” (Bodily Sensation), since most countries in Hispanic America have a warm or temperate climate year round. These four items were therefore removed and replaced with more contextually relevant items.

The item “newly cut grass” (Smell) was reworded to “damp grass”. In Hispanic America, grass tends to be found in communal parks or areas, but not in private gardens, and thus people rarely cut grass.

An alternative item pool for the subscales Taste, Smell and Bodily Sensations was developed and refined through group discussions among the research team. We retained only those items deemed most comprehensible and familiar to participants across Hispanic America from all socio-economic backgrounds. The new items included for testing were: “car exhaust fumes” (Smell); “mint leaves” (Taste); “orange juice” (Taste); “sitting on a hard wooden chair” (Bodily Sensation); and “your leg falling asleep” (Bodily Sensation).

The scale instructions were lengthened in detail to ensure that respondents understood what was being asked of them. Additionally, the response scale was edited to assist comprehension by (i) shading the scale in darkening shades of grey from left to right, and (ii) using additional word anchors to clarify the meaning of the scale values (see the final version at the end of this article for instructions and response scale).

Descriptive statistics of adapted PSI-Q

Table 2 presents means and standard deviations for the individual items, scale and subscale totals for the full sample of N = 508 who completed the adapted 36-item scale. The items with comparatively low vividness (M < 7) were: “a stuffy room” (Smell), “burnt wood” (Smell), “warm sand” (Touch), and “scared” (Feeling).

Values for Cohen’s alpha and McDonald’s omega, classic measures of internal reliability, reveal acceptable to good internal consistency, assuming unidimensionality, for all subscales of the PSI-Q (see Table 3). Consistency did not improve with the removal of any item, with the exception of “jumping into a pool”, which marginally improved reliability of Bodily Sensations to α = 0.75, ω = 0.76. Correlations between the subscales are moderate except for the Emotion subscale, which has the lowest correlations with the other dimensions (See Table 3).

The VVIQ-2 scale exhibited a good reliability (α =. 87, ω = 0.90). As before, moderate convergent validity was seen between the VVIQ-2 and the PSI-Q full scale (r=0.57). All PSI-Q subscale scores were positively associated with the VVIQ-2, with the Taste and Feeling subscales having the highest (r = .52) and lowest correlations (r = .39) respectively.

Confirmatory factor analysis

An initial confirmatory factor analysis showed an acceptable fit for replication, with an RMSEA of 0.07 and a SRMR of 0.067 (See Table 3). The model fit, however, improves significantly with various modifications, as outlined below.

Modification indices suggested fitting a model including residual covariances between items in the Feelings scale and between the items “car fumes” and “burnt wood”. This model, also presented in Table 3 (third column), is the best fitting model without allowing item cross-loading. However, this model still does not exhibit a good fit, with CFI and TLI indices under .90 and an RMSEA confidence interval that does not include 0.05.

The modification indices of the model fitted to the original structure suggest several cross-loadings, particularly of Feeling items (“furious”, “excited”, “scared”) onto the Touch and Bodily Sensations scales. While allowing for cross-loadings improves the fit of the simple structure imposed by the analysis, it limits the interpretation of the relevant scores as measures of the constructs at hand. Alternatively, shortening the scale can help to reduce the error variance (Eriksson & Boman, 2018) while maintaining the integrity of the subscales’ scores. We fitted a model with a reduced number of items, by deleting those with factor loadings < .35; specifically, from the subscales Touch (items “fur” and “soft towel”), Bodily Sensations (“jumping in a pool”), Smell (“rose”), Taste (“salty water”) and Feelings (“furious” and “scared”).

The indices for both the model with cross-loadings and the shortened 29-item version are presented in Table 3 (fourth and fifth column). Both models’ indices show an improvement relative to the original model and to the best-fitting model without cross-loadings, with the shortened model having the best fit (see Electronic Supplementary Material 2 for factor loadings for this model).

As a result, the version of the PSI-Q that best captures the underlying imaginal dimensions in a Hispanic American context is the Shortened Version, which exhibits the best overall fit (See Table 4), acceptable loading for all items included (See Electronic Supplementary material 2), and includes at least 3 items per dimension (included at the end of this article). The main difference with the original version lies in the measurement of the Feelings and Touch dimensions, whose items did not seem to uniquely capture the dimension intended.

Moderators of imagery vividness

To determine if there were significant moderators of imagery vividness, we fitted a linear model with age, gender (after eliminating non-responders), and educational level as predictors of the scores in all items. The model ( R2 = 0.014. F(7, 3527) = 8.35, p < 0.0001) revealed differences between men and women and an interaction between gender and education. Women scored slightly higher than men in overall scores, particularly in the Feelings (MMen = 7.90, MWomen = 8.16, Welch t (496) = -1.83, p = 0.06) and Taste subscales (MMen = 7.86, MWomen = 8.40, Welch t (483) = -3.66, p < 0.001). The interaction between gender and education revealed that the link between visualization and education is stronger for men than for women (t (3541) = -2.13, p = 0.021); in men (but not women) higher formal education is associated with stronger imagery vividness. No other terms or predictors were significant.