Verbal fluency tests: Normative data for Spanish-speaking pediatric population

2.1Participants

The sample consisted of 4,373 healthy children who were recruited from Chile, Cuba, Ecuador, Guatemala, Honduras, Mexico, Paraguay, Peru, Puerto Rico, and Spain. Participants were selected according to the following criteria: a) were between 6 and 17 years of age, b) were born and currently lived in the country where the study was conducted, c) Spanish as primary language, d) an IQ ≥80 on the Test of Non-Verbal Intelligence (TONI-2, Brown, Sherbenou, & Johnsen, 2009), and e) a score <19 on the Children’s Depression Inventory (CDI, Kovacs, 1992).

Children with history of neurologic or psychiatric disorders as reported by the participant’s parent(s), were excluded due to its effects on cognitive performance. Participants in the study were from public or private schools, and they signed an informed consent to participate. Socio-demographic and participant characteristics for each of the countries’ samples have been reported elsewhere (Rivera & Arango-Lasprilla, 2017). Ethics Committee approval was obtained for the study in each country.

2.2Instrument administration

VFT are commonly used in clinical practice and research for their sensitivity to brain damage. There are different modalities of VFT, but the two more commontly used are the phonological and semantic VFT. In phonological VFT, the participant is required to produce in 60 seconds as many words as s/he can beginning with certain letters (in this study F, A, and S). However, in semantic VFT, the participant is required to produce in 60 seconds as many words as s/he can belonging to a particular category (in this study, animals and fruits). In this process, cognitive flexibility, information organization, and inhibitory capacity are activated, so poor performance is indicative of dysfunction of the frontal lobe, specifically the left frontal cortex (Gouveia, Brucki, Malheiros, & Bueno, 2007). For administration and scoring, Olabarrieta-Landa, Landa-Torre, López-Mugartza, Bialystok, and Arango-Lasprilla (2017) guidelines were followed.

2.3Statistical analyses

Detailed statistical analyses used to generate the normative data for the VFT scores are described in Rivera & Arango-Lasprilla (2017). In summary, the scores were standardized using multiple linear regression analyses by means of a four-step procedure.

1) The letters F, A, S and the categories of animals and fruits scores were computed separately by means of the final multiple regression models. The full regression models included as predictors: age, age², sex, and mean level of parental education (MLPE). Age was centered (= calendar age – mean age in the sample by country) before computing the quadratic age term to avoid multicollinearity (Aiken & West, 1991). Sex was coded as male = 1 and female = 0. The MLPE variable was coded as 1 if the participant’s parent(s) had >12 years of education or 0 if participant’s parent(s) had ≤12 years of education. If predicted variables were not statistically significant in the multivariate model with an alpha of 0.05, the non-significant variables were removed and the model was run again. A final regression model was conducted: yˆi=B0+B1·(Age-x¯Agebycountry)i+B2·(Age-x¯Agebycountry)i2+B3·Sexi+B4·MLPEi . 2) Residual scores were calculated based on the final model ( ei=yi-yˆi ). 3) Residuals were standardized using the residual Standard Deviation (SD_e) value provided by the regression model: z_i = e_i/SD_e. 4) Standardized residuals were converted to percentile values using the standard normal cumulative distribution function. 4) Standardized residuals were converted to percentile values using the standard normal cumulative distribution function. This four-step process was applied to letters F, A, S and the categories of animals and fruits scores separately for each country.

For all multiple linear regression models, the following assumptions were evaluated: a) multicollinearity by the values of the Variance Inflation Factor (VIF), which must not exceed 10, and the collinearity tolerance values, which must not exceed the value of 1 (Kutner, Nachtsheim, Neter, & Li, 2005), and b) the existence of influential values by calculating the Cook’s distance. The maximum Cook’s distance value was related to a F (p, n - p) distribution. Influential values are considered when percentile value is equal or higher than 50 (Cook, 1977; Kutner et al., 2005). All analyzes were performed using SPSS version 23 (IBM Corp., Armonk, NY).

3.1Phonological fluency – letter F

The final multivariate linear regression models for the ten country-specific letter F score were significant (Table 1). In all countries, the letter F score increased linearly as a function of age. The letter F score for Honduras was also affected by a quadratic age effect. Children from Guatemala, Mexico and Spain who had parent(s) with a MLPE >12 years obtained higher letter F score than children whose parent(s) had a MLPE ≤12 years. The child’s sex affected letter F score for Mexico, Paraguay, Puerto Rico, and Spain, such that girls scored higher than boys. The amount of variance these predictors explained in the letter F score ranged from 26.0% (Guatemala) to 42.2% (Honduras).

3.2Phonological fluency – letter A

The final multivariate linear regression models for the ten country-specific letter A score were significant (Table 2). In all countries, the letter A score increased linearly as a function of age. The letter A score for Cuba was also affected by a quadratic age effect. Children from Chile, Mexico, Puerto Rico, and Spain who had parent(s) with a MLPE >12 years obtained higher letter A score than children who had parent(s) with a MLPE ≤12 years. The child’s sex affected letter A score for Ecuador and Paraguay such that girls scored higher than boys. The amount of variance these predictors explained in the letter A score ranged from 23.0% (Guatemala) to 49.3% (Chile).

3.3Phonological fluency – letter S

The final multivariate linear regression models for the ten country-specific letter S score were significant (Table 3). In all countries, the letter S score increased linearly as a function of age. The S letter score for Spain was also affected by a quadratic age effect. Children from Ecuador, Guatemala, Mexico, and Spain who had parent(s) with a MLPE >12 years obtained higher letter S score than children who had parent(s) with a MLPE ≤12 years. The child’s sex affected letter S score for Paraguay, such that girls scored higher than boys. The amount of variance these predictors explained in the letter S score ranged from 22.9% (Guatemala) to 44.4% (Honduras).

3.4Semantic fluency – animals

The final multivariate linear regression models for the ten country-specific animals’ category score were significant (Table 4). In all countries, the animal’s category score increased linearly as a function of age. The animals’ category score for Chile, Ecuador, Mexico, and Spain was also affected by a quadratic age effect. Children from Chile, Ecuador, Guatelama, Honduras, Mexico, Puerto Rico, and Spain who had parent(s) with a MLPE >12 years obtained higher animals’ category score than children who had parent(s) with a MLPE ≤12 years. The child’s sex affected animals’ category score for Chile and Puerto Rico. In Chile, boys achieved higher scores than girls; however, in Puerto Rico, females achieved higher scores than males. The amount of variance these predictors explained in the animals’ category score ranged from 24.3% (Paraguay) to 46.6% (Cuba).

3.5Semantic fluency – fruits

The final multivariate linear regression models for the ten country-specific fruit’s category score were significant (Table 5). In all countries, the fruit’s category score increased linearly as a function of age. The scores for Cuba, Ecuador, Mexico, Peru, and Spain were affected by a quadratic age effect. Children from Chile, Ecuador, Mexico, and Spain who had parent(s) with a MLPE >12 years obtained higher fruits’ category score than children who had parent(s) with a MLPE ≤12 years. The child’s sex affected fruits’ category score for Ecuador, Mexico, Paraguay, Puerto Rico, and Spain, such that girls achieved higher scores than boys. The amount of variance these predictors explained in the fruits’ category score ranged from 17.2% (Guatemala) to 46.2% (Cuba).

The assumptions of multiple linear regression analysis were met for all final models. There was not multicollinearity (the VIF values were below 10; VIF ≤1.050; collinearity tolerance values did not exceed the value of 1) or influential cases (the maximum Cook’s distance value was 0.117 in a F _(2,298)  distribution which correspond to percentile 11).

3.6Normative procedure

Norms (e.g., a percentile score) for the different VFT scores by country were established using the four-step procedure described in the statistical analysis section. An example is provided next to facilitate the understanding of the procedure used to obtain the percentile associated with a score on this test. To find the percentile score for an 8-year-old Mexican girl who scored a 15 on the animal’s category and whose parent(s) have a MLPE of 14, the steps are as follows:

1) Find Mexico in Table 4, which provides the final regression models by country for the animals’ category scores. Use the B weights to create an equation to obtain the predicted animals’ category score for this child using the coding provided in the statistical analysis section. The corresponding B weights are multiplied by the centered age (= calendar age – mean age in the Mexican sample which is equal to 11.4 years), centered age² (= calendar age – mean age in the Mexican sample which is equal to 11.4 years)², and MLPE code based on the 12 years of education threshold. See Rivera and Arango-Lasprilla (2017) to find the mean age of each country’s sample. Then the result is added to the constant generated by the model in order to calculate the predicted value. Child’s sex was not a significant predictor, and therefore is not included in this model.

In the case of our example of a Mexican girl, the predicted animals’ category score would be calculated using the following equation: yˆi=15.618+[0.842·(Agei-11.4)] + [(-0.083·(Agei-11.4)2]+(1.621·MLPE) . The girl’s age is 8. The MLPE (14 years) is split into either ≤12 years (and assigned a 0) or more than 12 years (and assigned a 1) in the model. Since the parent(s) of the hypothetical child in the example have 14 years of education, the MLPE value is 1. Thus, the predicted value equation is: yˆi=15.618+[0.842·(8-11.4)]+[-0.083·(8-11.4)2]+(1.621·1)=15.618+(-2.863)+(-0.959)+1.621=13.417.

2) To calculate the residual value (indicated with an e_i in the equation), subtract the actual animals’ category score (she scored 15) from the predicted value we just calculated (ei=yi-yˆi) . In this case, the residual would be e_i = 15 - 13.417 = 1.583. 3) Next, consult the SD_e column in Table 4 to obtain the country-specific SD_e (residual) value. For Mexico, it is 3.951. Using this value, the residual value can be transformed to a standardized z score using the equation z_i = e_i/SD_e. In this case, the z_i is 1.582/3.951 = 0.400. This is the standardized z score for an 8-year-old Mexican girl who scored a 15 on the animal’s category who has parent(s) with a MLPE of 14.

4) The last step is to use the tables available in most statistical reference books (e.g., Strauss et al., 2006) to convert z scores to percentiles. In this example, the z score (probability) of 0.400 corresponds to the 66^th percentile. It is important to remember to use the appropriate tables that correspond to each test (letters F, A, S and the categories of animals and fruits) when performing these calculations.

3.7User-friendly normative data

The four-step normative procedures explained above offers the clinician the ability to determine an exact percentile for a child who has a specific score on the VFT. However, this method can be prone to human error due to the number of required computations by hand. To enhance user-friendliness, the authors have completed these steps for a range of raw scores based on age, sex, and MLPE and created tables for clinicians to more easily obtain a percentile range/estimate associated with a given raw score on this test. These tables are available by country and type of test in the Appendix. To obtain an approximate percentile for the above example (i.e., converting a raw score of 15 on the animals’ category test for a Mexican girl who is 8 years old whose parent(s) have a MLPE of 14) using the simplified normative tables provided in the Appendix, the following steps must be followed. (1) First, identify the appropriate table ensuring the appropriate country and test (letters F, A, S and the categories of animals and fruits). In this case, the table for animals’ category score for Mexico can be found in Table A44. (2) Next, the table is divided based on MLPE (≤12 vs. more than 12 years of education). Since the parent(s) had 14 years of education, the upper section of the table for >12 years of MLPE will be used. (3) Find the appropriate age of the child, in this case, 8 years old. (4) Next, look in the 8 years’ age column to find the approximate location of the raw score obtained on the test. Within the 8 years’ column, the score of 15 obtained by this Mexican girl corresponds to an approximate percentile of 70.

The percentile obtained using this user-friendly table is slightly different than the hand-calculated, more accurate method (i.e., 66^th vs. 70^th) because the user-friendly table is based on a limited number of percentile values. Individual percentiles cannot be presented in these tables due to space limitations. If the exact score is not listed in the column, the percentile value must be estimated from the list of raw scores available.

4.1Limitations and future directions

The study has some limitations. First, only children whose primary language was Spanish were recruited. Therefore, the results of this study cannot be generalized to children whose primary language is not Spanish. This is especially important because many countries in Latin America and Spain are multilingual, with Spanish being one of the most spoken languages among its inhabitants. For example, in Mexico there are 62 indigenous languages (Chamoreau, 2014), and in Spain there are other official languages apart from Spanish such as Catalan, Basque or Galician (Garrido Medina, 2007).

Second, the results of this study cannot be generalized to other Latin American countries such as Argentina, Bolivia, Panama, or Venezuela, among others. Future studies should obtain normative data for VFT in these countries. Third, even though in Chile, Mexico, Paraguay, Puerto Rico, and Spain the sample was collected from several geographic areas of the country, for the rest of the countries of the study only children from a specific geographic area were recruited. It would have been ideal to recruit children from different areas of these countries as well. Future studies should expand the data from this study with samples from different geographic regions. Likewise, most of the children were recruited from urban areas, thus future studies should include more children from rural areas.

Fourth, the letters selected for the study were F, A, and S. These letters are commonly used by professionals in Latin America. However, other letters have been proposed to be used with Spanish-speaking populations, so future studies should also evaluate whether performance changes depending on letters selected.

Lastly, normative data were generated using healthy children and adolescents. Future studies should include clinical samples (e.g., children and adolescents with brain damage, epilepsy, etc.) to obtain proper cutoff points and calculate the sensitivity and specificity of VFT.

4.2Implications and conclusions

Despite the limitations, this is the first study to offer normative data for the phonological and semantic VFT in nine countries in Latin America and Spain among 4,373 children and adolescents. Most normative data studies have focused on adult populations, so to date this represents the largest normative data study of neuropsychological tests for children and adolescents in the world. In addition, for the creation of the normative data, more up-to-date approaches have been used to allow a more accurate calculation of the norms, such as the use of multiple regressions and residual values. Therefore, these norms are expected to be useful for those who work in the neuropsychological evaluation of children and adolescents in these 10 countries.