Verbal fluency tests: Normative data for the Latin American Spanish speaking adult population

2.1Participants

The sample consisted of 3,977 healthy individuals who were recruited from Argentina, Bolivia, Chile, Cuba, El Salvador, Guatemala, Honduras, Mexico, Paraguay, Peru, and, Puerto Rico. The participants were selected according to the following criteria: a) were between 18 to 95 years of age, b) were born and currently lived in the country where the protocol was conducted, c) spoke Spanish as their native language, d) had completed at least one year of formal education, e) were able to read and write at the time of evaluation, f) scored ≥23 on the Mini-Mental State Examination (MMSE, Folstein, Folstein, & McHugh, 1975), g) scored ≤4 on the Patient Health Questionnaire–9 (PHQ-9, Kroenke, Spitzer, & Williams, 2001), and h) scored ≥90 on the Barthel Index (Mahoney & Barthel, 1965).

Participants with self-reported neurologic or psychiatric disorders were excluded due to a potential effect on cognitive performance. Participants were volunteers from the community and signed an informed consent. Nine participants were excluded from the analyses, with a final sample of 3,961 participants. Socio-demographic and participant characteristics for each of the countries’ samples have been reported elsewhere Guárdia-Olmos, Peró-Cebollero, Rivera, & Arango-Lasprilla, (2015). The multi-center study was approved by the Ethics Committee of the coordinating site, the University of Deusto, Spain.

2.2Instrument administration

For the present study, phonological and semantic verbal fluency tests were applied. The aim of each test is to create, within 60 seconds, as many words as participants can that begin with certain letter (F, A, S) or that belong to a particular category (animals). Participants were told to avoid proper names, augmentatives, and diminutives. At the same time, the examiner provided prompts if the participant gave no responses within 10 seconds during each trial. The total score consisted of the total correct answers for each letter or category.

2.3Statistical analyses

The detailed statistical analyses used to generate the normative data for this test are described in Guárdia-Olmos et al. (2015). In summary, the data manipulation process for each country-specific dataset involved five-steps: a) t –tests for independent samples and effect sizes (r) were conducted to determine gender effects. If the effect size was larger than 0.3, gender was included in the model with gender dummy coded and female as the reference group (male = 1 and female = 0). b) A multivariable regression model was used to specify the predictive model including gender (if effect size was larger than 0.3), age as a continuous variable, and education as a dummy coded variable with 1 if the participants had >12 years of education and 0 if participants had 1–12 years of education. If gender, age and/or education were not statistically significant in this multivariate model with an alpha of 0.05, the non-significant variables were removed and the model was re-run. Then a final regression model was conducted that included age (if statistically significant in the multivariate model), dichotomized education (if statistically significant in the multivariate model), and/or gender (if effect size was greater than 0.3) [yˆi=β0+(βAge.Agei)+(βEduc.Educi)+(βGender.Genderi)]; c) residual scores were calculated based on this final model (ei=yi-yˆi); d) using the SD (residual) value provided by the regression model, residuals were standardized: z = e _i /SD _e , with SD _e (residual) = the standard deviation of the residuals in the normative sample; and e) standardized residuals were converted to percentile values (Strauss et al., 2006). Using each country’s dataset, these steps were applied to phonological and semantic verbal fluency scores.

3.1Phonological fluency - letter F

Regarding the effect of gender on letter F, the t-tests showed significant differences between men and women from Guatemala, however, its effect size was less than 0.3. Table 1 shows the results of the gender analyses by country on letter F. As shown in Table 1, the effect sizes for all countries were less than 0.3, and therefore gender was not taken into account to generate letter F normative data for any of the countries in the study.

The final eleven letter F multivariate linear regression models for each country are shown in Table 2. In all countries, except Argentina, Guatemala, Honduras, and, Paraguay, the letter F score increased for those with more than 12 years of education (see Table 2) and, in all countries except Argentina, Guatemala, Honduras, and, Paraguay, letter F score decreased in a linear fashion as a function of age. The amount of variance explained in letter F scores ranged from 8% (in Cuba) to 30% (in Chile).

3.2Phonological fluency - letter A

Regarding the effect of gender on letter A, the t-tests showed significant differences between men and women in the countries of Chile, Guatemala, and, Paraguay, however, none of these three countries had an effect size larger than 0.3. Table 3 shows the results of the gender analysis by country on letter A. As shown in Table 1, the effect sizes for all countries were less than 0.3, and therefore gender was not taken into account to generate letter A normative data for any of the countries in the study.

The final eleven letter A multivariate linear regression models for each country are shown in Table 4. In all countries, the letter A score increased for those with more than 12 years of education (see Table 4) and decreased in a linear fashion as a function of age except in Argentina, Guatemala and, Paraguay. The amount of variance explained in letter A scores ranged from 7% (in Paraguay) to 32% (in Chile).

3.3Phonological fluency - letter S

Regarding the effect of gender on letter S, the t-tests showed significant differences between men and women from Puerto Rico, however, its effect size was less than 0.3. Table 5 shows the results of the gender analysis by country on letter S. As shown in Table 5, the effect sizes for all countries were less than 0.3, and therefore gender was not taken into account to generate letter S normative data for any of the countries in the study.

The final eleven letter S multivariate linear regression models for each country are shown in Table 6. In all countries, the letter S score increased for those with more than 12 years of education (see Table 6) and decreased in linear fashion as a function of age except in Argentina, Guatemala and, Paraguay. The amount of variance explained in letter S scores ranged from 8% (in Cuba) to 32% (in Chile).

3.4Semantic fluency - animals

Regarding the effect of gender on animals, the t-tests showed significant differences between men and women in Guatemala, however, its effect size was less than 0.3. Table 7 shows the results of the gender analysis by country on animals. As shown in Table 7, the effect sizes for all countries were less than 0.3, and therefore gender was not taken into account to generate animals normative data for any of the countries in the study.

The final eleven animals multivariate linear regression models for each country are shown in Table 8. In all countries, the animals score increased for those with more than 12 years of education (see Table 2) and decreased in a linear fashion as a function of age. The amount of variance explained in animals scores ranged from 16% (in Cuba) to 43% (in Paraguay).

3.5Normative procedure

Norms (e.g., a percentile score) for the verbal fluency different scores were established using the five-step procedure described above. To facilitate the understanding of the procedure to obtain the percentile associated with a score on this test, an example will be given. Suppose you need to find the percentile score for a Mexican woman, who is 50 years old and has 8 years of education. She has a score of 20 on animals. The steps to obtain the percentile for this score are: a) Check Table 7 to determine if the effect size of gender in the country of interest (Mexico) on this test and time point (animals) is greater than 0.3 by country. The column labelled r in Table 7 indicates the effect size. In this example, the effect size is 0.034, which is not greater than 0.3. For Mexicans on this test, gender does not influence scores to a sufficient degree to take it into account when determining the percentile. b) Find Mexico in Table 8, which provides the final regression models by country for animals. Use the B weights to create an equation that will allow you to obtain the predicted animals score. The corresponding B weights are multiplied by the actual age and dichotomized education scores and added to a constant in order to calculate the predicted value. In this case, the predicted animals score would be calculated using the equation [yˆi=21.551+(-0.087·Agei)+(2.043·DichotomizedEducationalLeveli)] (the values have been rounded for presentation in the formula).The subscript notation i indicate the person of interest. The person’s age is 50, but the education variable is not continuous in the model. Years of education is split into either 1 to 12 years (and assigned a 0) or more than 12 years (and assigned a 1) in the model. Since our hypothetical person in the example has 8 years of education, her educational level value is 0. Thus the predicted value is yˆi=21.551+(-0.087.50)+(2.043.0)=21.551-4.350+0=17.201). c) In order to calculate the residual value (indicated with an e in the equation), we subtract the actual value from the predicted value we just calculated (ei=yi-yˆi). In this case, it would be e _i  = 20 - 17.201 = 2.80. d) Next, consult the SD _e column in Table 8 to obtain the country-specific SD _e (residual) value. For Mexico it is 4.673. Using this value, we can transform the residual value to a standardized z score using the equation (e _i /SD _e ). In this case, we have (2.80)/4.673 = 0.599. This is the standardized z score for a Mexican woman aged 50 and 8 years of education and a score of 20 on animals. e) The last step is to look-up the tables in the statistical reference books (e.g. Strauss et al., 2006) or use a trusted online calculator like the one available at http://www.measuringu.com/pcalcz.php. In the online calculator, you would enter the z score and choose a one-sided test and note the percent of area after hitting the submit button. In this case, the probability of 0.599 corresponds to the 72th percentile. Please remember to use the appropriate tables that correspond to each test (letter vs. animal) when performing these calculations. If the percentile for the others verbal fluency scores is desired, Tables 1–6 must beused.

3.6User-friendly normative data

The five-step normative procedures explained above can provide more individualized norms. However, this method can be prone to human error due to the number of required computations. To enhance user-friendliness, the authors have completed these steps for a range of raw scores based on small age range groupings (see Guárdia-Olmos et al., 2015) and created tables so that clinicians can more easily use to obtain a percentile range associated with a given raw score on this test. These tables are available by country and type of test in the Appendix. In order to obtain an approximate percentile for the above example (converting a raw score of 20 in animals for a Mexican women who is 50 years old and has 8 years of education using the simplified normative tables provided, the following steps are recommended. (1) First, identify the appropriate table ensuring the specific country and test. In this case, the table for animals scores for Mexico can be found in Table A41. (2) Note if the title of the table indicates that it is only to be used for one specific gender. In this case, gender is not specified. Thus Table A41 is used for both males and females. (3) Next, the table is divided based on educational level (1 to 12 vs. 12 years of education). Since this woman has 8 years of education, she falls into 1–12 years of education category. These data can be found in the top section of the table. (4) Determine the age range most appropriate for the individual. In this case, 50 falls into the column 48–52 years of age. (5) Read down the age range column to find the approximate location of the raw score the person obtained on the test. Reading down the 48–52 column, the score of 20 obtained by this Mexican woman corresponds to an approximate percentile of 70.

The percentile obtained via this user-friendly table method (70th) is slightly different than the more exact one (72th) obtained following the individual conversion steps above because the table method is based on an age range (e.g., individuals aged 48–52) instead of the exact age (individuals aged 50). If the exact score is not listed in the column, you must estimate the percentile value from the listed raw scores.

4.1Limitations and future directions

The current study has several limitations and directions for future research. First, participants in this study spoke Spanish as a primary language, and no data were collected on bilingualism. Verbal fluency test performance could likely differ among individuals who speak secondary languages, so future research should examine effects of bilingualism on verbal fluency test performance. Participants were recruited in specific cities and regions of the countries in this study as opposed to nationally. Although the current study is the largest verbal fluency test normative study to date in Latin America, or in any global region, it should be seen as a first step toward larger, nationally representative normative studies. Although many participants in this study had fewer than 12 years of education, illiterate individuals did not participate, and the current norms cannot generalize well to illiterate adults. Similarly, participants with a history of neurological conditions and children were excluded, so future normative research should be conducted with these groups.

Second, neuropsychologists need to exercise caution in using the verbal fluency test norms from this study for people in countries outside of those from which data were collected. Researchers need to create verbal fluency test norms in Latin American countries such as Ecuador, Uruguay, Venezuela, and Panama. Despite this limitation, these verbal fluency test norms may be actually more accurate in Latin American countries not a part of this study than current norms in use, but this generalizability needs to be investigated in future studies.

Third, the verbal fluency tests are common neuropsychological instruments in Latin America, but other measures need to be normed using a similar approach to improve their accuracy in this region. Future research should also examine the psychometrics and ecological validity of the verbal fluency tests, and that of other common assessments in Latin America. Researchers need to create instruments within Latin American cultures with good ecological validity, as the verbal fluency tests were created in a Western culture perhaps very different from the diverse cultures in Latin America. Future research can develop assessments within local cultures, not simply translate and norm tests from other countries.

Despite these limitations, limited studies have produced verbal fluency test norms in Spanish-speakers in the United States (Lucas et al., 1998; Stricks et al., 1998; Acevedo et al., 2000). The current study was the first to generate verbal fluency test norms across 11 countries in Latin America with nearly 4,000 participants. As a result, it was the largest, most comprehensive verbal fluency test normative study to date in any global region, and its norms will influence the standard of neuropsychological assessment with the verbal fluency tests in Latin America unlike any study before it.