TYPE
Original Research
10 October 2022
10.3389/fpsyt.2022.973134
PUBLISHED
DOI
OPEN ACCESS
EDITED BY
Michael Noll-Hussong,
Saarland University, Germany
REVIEWED BY
Alejandra Dominguez Espinosa,
Ibero American University, Mexico
Marinés Mejía Alvarez,
AIGLÉ Foundation, Argentina
*CORRESPONDENCE
Filiberto Toledano-Toledano
filiberto.toledano.phd@gmail.com
†
These authors have contributed
equally to this work and share first
authorship
SPECIALTY SECTION
This article was submitted to
Anxiety and Stress Disorders,
a section of the journal
Frontiers in Psychiatry
19 June 2022
09 September 2022
PUBLISHED 10 October 2022
RECEIVED
ACCEPTED
CITATION
Astudillo-García CI, Austria-Corrales F,
Rivera-Rivera L,
Reynales-Shigematsu LM,
Gómez-García JA, Séris-Martinez M,
Jiménez-Tapia A, Robles R,
Morales-Chainé S, López-Montoya A,
Cuevas-Renaud C and
Toledano-Toledano F (2022)
Measurement invariance of the GAD-5
Generalized Anxiety Disorder Scale in a
Mexican general population sample.
Front. Psychiatry 13:973134.
doi: 10.3389/fpsyt.2022.973134
COPYRIGHT
© 2022 Astudillo-García,
Austria-Corrales, Rivera-Rivera,
Reynales-Shigematsu, Gómez-García,
Séris-Martinez, Jiménez-Tapia, Robles,
Morales-Chainé, López-Montoya,
Cuevas-Renaud and
Toledano-Toledano. This is an
open-access article distributed under
the terms of the Creative Commons
Attribution License (CC BY). The use,
distribution or reproduction in other
forums is permitted, provided the
original author(s) and the copyright
owner(s) are credited and that the
original publication in this journal is
cited, in accordance with accepted
academic practice. No use, distribution
or reproduction is permitted which
does not comply with these terms.
Measurement invariance of the
GAD-5 Generalized Anxiety
Disorder Scale in a Mexican
general population sample
Claudia I. Astudillo-García1† , Fernando Austria-Corrales2† ,
Leonor Rivera-Rivera3 , Luz Myriam Reynales-Shigematsu3 ,
José Alberto Gómez-García4 , Marina Séris-Martinez3 ,
Alberto Jiménez-Tapia5 , Rebeca Robles5 ,
Silvia Morales-Chainé6 , Alejandra López-Montoya6 ,
Corina Cuevas-Renaud6 and Filiberto Toledano-Toledano7,8,9*
1
Servicios de Atención Psiquiátrica (SAP), Secretaría de Salud, Ciudad de México, México, 2 Comisión
Nacional para la Mejora Continua de la Educación (MEJOREDU), Ciudad de México, México,
3
Centro de Investigación en Salud Poblacional, Instituto Nacional de Salud Pública (INSP),
Cuernavaca, Morelos, México, 4 Secretariado Técnico del Consejo Nacional de Salud Mental
(STCONSAME), Secretaría de Salud, Ciudad de México, México, 5 Instituto Nacional de Psiquiatría
Ramón de la Fuente Muñiz (INPRFM), Ciudad de México, México, 6 Facultad de Psicología,
Universidad Nacional Autónoma de México, Ciudad de México, México, 7 Unidad de Investigación en
Medicina Basada en Evidencias, Hospital Infantil de México Federico Gómez,Instituto Nacional de
Salud, Ciudad de México, Mexico, 8 Unidad de Investigación Sociomédica, Instituto Nacional de
Rehabilitación Luis Guillermo Ibarra Ibarra, Ciudad de México, México, 9 Dirección de Investigación y
Diseminación del Conocimiento, Instituto Nacional de Ciencias e Innovación para la Formación de
Comunidad Científica, INDEHUS, Ciudad de México, México
The primary objective of this study was to evaluate the measurement
of invariance by sex, age, and educational level of an online version of
the Generalized Anxiety Disorder Scale in a five-item version (GAD-5).
Configural, metric, scalar, and strict invariance were evaluated using data
from 79,473 respondents who answered a mental health questionnaire
during the COVID-19 pandemic in Mexico. The sex variable was classified
as male or female; age was categorized as minors, youth, young adults,
adults, and older adults; and educational level was divided into basic, upper
secondary, higher, and graduate education. To test for configural invariance,
confirmatory factor models were constructed. For metric invariance, equality
restrictions were established for the factor loadings between the construct
and its items; for scalar invariance, equality restrictions were established
between the intercepts; strict variance implied the additional restriction of
the residuals. Statistical analysis was performed in R software with the lavaan
package. The results show that with respect to sex, age, and educational level,
configural and metric measurement invariance was confirmed (1CFI < 0.002;
1RMSEA < 0.015). However, with respect to scalar and strict invariance, the
results showed significant differences regarding the fit model (1CFI > 0.002;
1RMSEA > 0.015). We conclude that the GAD-5 presents configural and metric
invariance for sex, age, and educational level, and scalar invariance for sex and
age groups. However, the scale does not demonstrate strict invariance. We
discuss the implications and suggest that this result could be related to the
evaluation of sociodemographic variables.
KEYWORDS
anxiety, Generalized Anxiety Disorder Scale (GAD), measurement invariance, multiplegroup analysis, factor analysis, statistical, mass screening
Frontiers in Psychiatry
01
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
Introduction
The factorial invariance of a scale is the statistical property
that indicates whether it measures the same latent construct
among the subgroups of a sample, which is a prerequisite
for making valid group comparisons. The presence of nonvariance could be indicative of bias due to differences in the
interpretation of the items included in a scale (9). To determine
whether a measure presents factorial invariance, factor loadings,
intercepts, and residual variances are tested to ensure that they
are equivalent in a factorial model that evaluates a latent concept.
To this end, a set of increasingly restricted structural equation
models are run to test whether differences between these models
are significant (10). Failure to test for invariance means that
different groups or subjects may respond differently to the items
and that factor means cannot be reasonably compared (10).
The GAD Scale was developed as a screening tool for
primary care settings (11). Its initial version consisted of nine
items reflecting all of the DSM-IV diagnostic criteria for the
disorder, as well as four items based on a review of existing
anxiety scales (11). A seven-item version (GAD-7) has reported
good to excellent sensitivity and specificity for most of the
relevant DSM-5 disorders (5) in both the general population
and in primary care patients (12). Measures of invariance have
been reported for the GAD-7 (6, 9, 13), but not for the GAD5, a five-item version obtained from studies of the primary care
population (3, 8). The five items are directly linked to the ICD11 diagnostic guidelines for depression and anxiety, in which
a total score of 3 or more predicted 89.6% of above-threshold
cases with generalized anxiety (11). This brief assessment of
anxiety minimizes the time required in the patient encounter
and obviates the need for paper and pencil tests and instrument
scoring (3). It therefore offers a substantially more practical
alternative for implementation in low-resource settings, and it
may also be of considerable value in high-income countries (3).
The confirmation of parameter invariance helps to verify
that the items and measures are free of biases that produce
differences, which could be the result of differences in age,
gender, and educational level. For example, the use of certain
words may create a difference between those who fully
understand an item and those who do not. In addition, gender
bias in the wording of items can generate systematic error
variances that may affect measurement precision. Confirming
the invariance of parameters across different ages, sexes, and
educational levels will help to understand whether the five
attributes measured by the GAD-5 are relatively constant
across groups and whether the groups analyzed share the same
metric: whether the construct being measured is equivalent
across groups (14). The aim of this study was thus to assess
measurement invariance through the estimates of configural,
metric, scalar, and strict invariance of the five-item version of
the Generalized Anxiety Disorder Scale (GAD-5), across sex, age
group, and educational level.
Anxiety disorders account for a large proportion of the
global burden of disease and disability. A systematic review
published in 2022 (1) reported that 301.4 million people
worldwide had some type of anxiety disorder, with an agestandardized prevalence rate of 3779.5 (3181.1–4473.3) per
1,00,000 population. However, in Latin America and the
Caribbean, this rate is 5502.3 (4625.9–6588.7). The global
prevalence of generalized anxiety disorder (GAD) was 4.5%
in 2021; although a higher prevalence has been reported in
high-income countries (5.3%) than in low-income countries
(2.8%), the proportion of people who have received treatment
is lower in the latter (19.2 vs. 38.4%) (2). In low- and middleincome countries, most people with these disorders will never
see a mental health specialist (3). It has also been reported that
subthreshold anxiety disorders may have twice the frequency
of the full syndrome, and are more persistent, cause greater
suffering and functional impairment, and have a higher risk of
onset and aggravation of other mental health conditions, such as
pain and comorbid somatic disorders, increasing care costs (4).
The existing differences by sex and age must be added
to this care gap. Women present greater anxiety than men.
According to the 2022 GBD review, 187.5 million women suffer
from anxiety disorders vs. 109.3 million men, in addition to the
fact that the number of disability-adjusted life years (DALYs)
increases steadily during childhood and adolescence, reaching
a maximum between the ages of 25 and 34 and decreasing
steadily after the age of 35 (1). In contexts such as the COVID-19
pandemic, evidence shows that there are significant differences
by sex and age, with women and younger people scoring
significantly higher in anxiety, and these differences are present
also by educational level (5). In order to make judgments across
conditions of age, sex, or educational level, scales are needed that
operate equivalently for these different groups of interest (6), and
that are available in non-specialized care settings.
Primary care is the ideal setting for the identification and
appropriate treatment of the most common mental disorders.
Screening for their early detection and treatment in primary
care can improve quality of life, help contain health care
costs, and limit complications from medical and mental health
comorbidities (7). The application of screening scales is a
useful alternative in primary care in low- and high-income
countries, given existing time and resource pressures (8). These
scales have the potential to improve case detection through
procedures that could be incorporated into primary care
practice. They direct attention to anxiety symptoms, and help
to determine the current status of the individual and offer a
specific diagnosis and treatment (8). Population-based screening
requires that such tools have psychometric properties that allow
for valid comparisons.
Frontiers in Psychiatry
02
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
Materials and methods
response form was modified to match the rest of the instruments
used in order to avoid having to provide different instructions
and response options for each part of the questionnaire. The
response options for the entire survey were a 10-point Likert
scale, where 0 indicated “does not describe me” and 10 “describes
me exactly.” With five items, the range of possible scores was
thus 0–50 points. There is evidence suggesting that increasing
the number of response options increases validity coefficients
by 0.04 (17). This evidence also suggests that the coefficients do
not rise artificially as the number of response options increases;
however, the validity does consistently improve. In another
study, Alwin (18) conducted a confirmatory factor analysis to
compare the performance of the versions with seven and eleven
response options and found that the latter had better validity and
reliability and lower invalidity indices.
Participants and procedure
We used a convenience sampling strategy to recruit 79,473
people who were analyzed for this study. Participants answered
the GAD-5 questionnaire from April 1 to December 31, 2020,
as part of the survey Atención Psicológica a Distancia para
la Salud Mental por la contingencia por COVID-19 (Remote
Mental Health Care during the COVID-19 Pandemic). This
survey was part of the Mexican effort, led by the Secretary
of Health, the Universidad Nacional Autónoma de México
(UNAM), the Instituto Nacional de Psiquiatría, and civil society
organizations to meet the mental health needs of the population
and reduce the stress caused by the pandemic. The survey was
administered by a team from the UNAM Faculty of Psychology
through the federal government’s coronavirus.gob.mx website.
On this website, people were invited to participate voluntarily
and confidentially and offered care resources according to the
risk levels detected for different mental health problems. The
questionnaire was self-administered online. A description of
the survey and the variables assessed is available in a previous
publication (15).
Data analysis
We first performed a confirmatory factor analysis (CFA),
using R software and the lavaan package (19), to test the
theoretical structure of the scale as well as its unidimensionality.
The covariance matrix was analyzed using the maximum
likelihood method, applying the Satorra–Bentler correction (20),
since the data do not assume multivariate normality. The fit of
the model was assessed with four fit indices. The comparative
fit index (CFI) takes possible values between 0 and 1, with a
value of at least 0.90 denoting adequate fit and a value greater
than or equal to 0.95 a very good fit. The Tucker-Lewis index
(TLI) also has a range from 0 to 1 with the same interpretation
criteria. The Root Mean Square Error Approximation (RMSEA)
should ideally have values of <0.06, although values of 0.08
are considered acceptable. Finally, the Standardized Root Mean
Square Residual (SRMR) is considered acceptable with a value
<0.10 and a good fit with a value <0.05 (21).
We next assessed measurement invariance using multigroup confirmatory factor analysis; this technique makes it
possible to gradually impose restrictions in order to test different
levels of parameter invariance: configural, metric, scalar, and
strict. The first step was to test configural invariance; this model
was used as a baseline for comparison with models that gradually
incorporated more equality constraints. To assess configural
invariance, it was necessary to keep the factor loading structure
constant between the different comparison groups, although
the values of the loadings, factor variances, and covariances
could vary because they were not restricted to being equal.
Metric invariance was subsequently determined by establishing
equality restrictions on the values of the factor loadings. We then
proceeded to test scalar invariance through the establishment
of equality restrictions between the intercepts, and finally strict
invariance, where equality was also restricted among residuals.
We evaluated changes in the comparative fit index (CFI) to
assess the measurement invariance between the different groups:
Study variables
The sociodemographic variables considered were sex, age
group, and educational level. Sex was classified as male or female.
Age was categorized as minors (13–17 years), youth (18–25
years), young adults (26–35 years), adults (36–59 years), and
older adults (60 years and older). Educational level was divided
into basic (elementary and junior high school), upper secondary
(high school or equivalent), higher education (undergraduate
degree) and graduate (specialty, master’s, and doctoral degrees).
The age categories are consistent with Mexican law that
considers adulthood to begin at age 18 and senior citizens
to be those over 60. The intermediate ages were divided into
three groups that represent the life trajectories of adults in
Mexico. However, it should be noted that the complexity of life
trajectories makes it difficult to construct a universal division
of different life stages (16). The categories of educational level
were based on the organization of the educational system
in Mexico, which includes basic (elementary and junior high
school), middle (high school), and higher education (university);
the latter was divided into separate categories for undergraduate
and graduate education.
The GAD-5 consists of five items: “I feel nervous, anxious, or
about to burst,” “I have felt unable to control my worries,” “I have
felt so worried, I have been unable to keep still,” “I have found it
hard to relax,” and “I have felt afraid that something terrible was
going to happen.” Participants were asked to what extent each
of these items described them in the past 2 weeks. The standard
Frontiers in Psychiatry
03
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
TABLE 1 Sociodemographic characteristics of the sample.
Variable
n (%)
Sex
Women
48,308 (60.79)
Men
31,165 (39.21)
Age group
Minors
6,392 (8.04)
Youth
14,967 (18.83)
Young adults
22,267 (28.02)
Adults
32,760 (41.22)
Older adults
3,087 (3.89)
Educational level
Basic education
11,703 (14.73)
Upper secondary education
23,444 (29.50)
Higher education
35,318 (44.44)
Graduate
9,008 (11.33)
FIGURE 1
Results of the confirmatory factor analysis of the GAD-5.
TABLE 2 Factor loads of GAD-5 items.
Item
of items in the GAD-5. The resulting model, as well as the
standardized parameters, can be seen in Figure 1.
Standardized
coefficient (β)*
I feel nervous, anxious, or about to burst
0.910
I have felt unable to control my worries
0.919
I have felt so worried I have been unable to keep still
0.865
I have found it hard to relax
0.899
I have felt afraid that something terrible was going to happen
0.823
Multi-group CFA and measurement
invariance
Once the unidimensionality of the GAD-5 and its parametric
stability were demonstrated, variances were divided by sex, age
group, and educational level, according to the categorizations
described above. Equality restrictions were then gradually
imposed, using the configural model as the baseline.
As regards invariance by sex, the configural invariance
showed a good fit with respect to the general model, indicating
a lack of significant differences in the factorial structure between
women and men. When equality restrictions were placed on
the factor loadings (metric invariance), no differences were
observed in the comparative fit index (1CFI = 0.000). This
evidence suggests that the GAD-5 is metrically invariant by
sex. Equality restrictions were then imposed on the intercepts
(scalar invariance), reducing the 1CFI by −0.001, suggesting a
lack of significant differences. Finally, after imposing equality
restrictions on residuals (strict invariance), a change of
−0.008 was observed in the 1CFI, a value of <0.01, the
traditional criterion for assessing the invariance of parameters.
As regards age, five groups were compared: minors, youth,
young adults, adults, and older adults. Table 3 shows that
differences in the 1CFI in the metric, scalar, and strict
invariance are in all cases less than the criteria established
by Cheung and Rensvold (22), suggesting that the GAD-5
is invariant at the configural, metric, scalar, and strict levels.
In relation to educational level, we observed that changes
in the 1CFI in metric, scalar, and strict invariance do not
* All values are significant, p < 0.001.
a change in CFI of −0.01 or more from the baseline was used
to reject the between-group invariance hypothesis (22). We also
evaluated 1SRMR and 1RMSEA as alternative fit indices, as
suggested by Chen (23).
Results
Data were analyzed from 79,473 people who participated
voluntarily and answered the questionnaire. The sample
included 60.79% women and 39.21% men, with an average age
of 35.11 years (SD = 12.74). The distribution by age group and
educational level is shown in Table 1.
GAD-5 factor analysis
The resulting model showed an adequate fit between the
theoretical model and the empirical data, as shown by the
following fit indices: CFI = 0.993; TLI = 0.987; RMSEA = 0.07,
CI [0.075, 0.081]; SMRM = 0.009. Table 2 shows the factor loads
Frontiers in Psychiatry
04
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
TABLE 3 Results of tests of measurement invariance.
Model
X2
df
CFI
TLI
SRMR
997.856
10
0.993
0.987
0.010
RMSEA
Model comparison
1CFI
By sex
Configural
0.079
-
-
(0.074-0.083)
Metric
1,223.088
14
0.993
0.990
0.013
0.068
Configural – Metric
0.000
(0.065-0.071)
Scalar
1,659.291
18
0.992
0.991
0.015
0.065
Metric – Scalar
−0.001
Scalar – Strict
−0.008
(0.063-0.068)
Strict
3,180.108
23
0.984
0.986
0.020
0.082
(0.080-0.085)
By age
Configural
1,126.013
25
0.993
0.986
0.010
0.082
-
-
(0.078-0.086)
Metric
1,646.677
41
0.992
0.990
0.020
0.069
Configural – Metric
−0.001
Metric – Scalar
−0.001
Scalar – Strict
−0.008
(0.066-0.072)
Scalar
2,190.414
57
0.991
0.992
0.022
Strict
3,165.974
77
0.983
0.989
0.023
0.063
(0.063-0.065)
0.073
(0.071-0.075)
By educational level
Configural
959.343
20
0.994
0.987
0.009
0.077
-
-
(0.073-0.081)
Metric
1,194.531
32
0.994
0.992
0.011
0.061
Configural – Metric
0.000
(0.058-0.064)
Scalar
2,054.791
44
0.991
0.991
0.017
0.063
Metric – Scalar
−0.003
Scalar – Strict
−0.005
(0.061-0.066)
Strict
2,713.069
59
0.986
0.990
0.019
0.067
(0.065-0.069)
exceed the −0.01 criterion, suggesting that the GAD-5 is
invariant across educational levels. The results are shown in
Table 3.
To confirm these results based on the traditional criteria
for assessing the invariance of parameters, the change in CFI
(1CFI), additional assessments were made using two alternative
indices suggested by Chen (23): changes in the RMSEA of 0.015
and the SMRM of 0.030 for metric invariance, and changes in the
scalar and strict invariance of 0.015. The results are summarized
in Table 4 for each of the comparison variables: sex, age group,
and educational level.
The results by sex and age showed that 1SRMR and
1RMSEA have values of <0.030 and 0.015 respectively
in assuming metric and scalar invariance, suggesting that
these invariances might be present, but not strict invariance.
However, the values observed for 1RMSEA indicate significant
differences in the model, so this possibility is not empirically
supported. As for educational level, there is only metric,
not scalar or strict invariance, since the 1RMSEA value
is −0.015.
Frontiers in Psychiatry
Taken together, these findings suggest that the GAD-5 has
psychometric properties that provide invariant measurements
for the sociodemographic characteristics of sex, age, and
educational level. However, the invariance is not complete in all
cases. The traditional 1CFI and alternative indexes of 1SRMR
and 1RMSEA coincide to show the following: (a) by sex, GAD5 has configural, metric, and scalar invariance; (b) by age group,
it has configural, metric, and scalar invariance; and (c) by
educational level, it has configural and metric invariance.
Discussion
Using data drawn from a large Mexican general population
sample, we assessed measurement invariance of the GAD-5
by sex, age, and educational level. Our findings indicate that
the GAD-5 conforms to the proposed theoretical structure,
since a unidimensional construct of generalized anxiety
symptomatology was obtained, which presented configural and
metric invariance in the comparison by sex, age, and educational
05
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
TABLE 4 Alternative fit indices to evaluate measurement invariance by sex, age, and education.
Model
SRMR
RMSEA
Model comparison
1CFI
1SRMR
1 RMSEA
-
-
-
Configural – Metric
−0.001
0.004
−0.011
Metric – Scalar
−0.001
0.002
−0.002
Scalar – Strict
−0.008
0.004
−0.017
-
-
-
0.000
0.011
−0.013
Metric – Scalar
−0.001
0.001
−0.006
Scalar – Strict
−0.008
0.002
0.010
By sex
Configural
0.010
0.079
-
(0.074–0.083)
Metric
0.013
0.068
(0.065–0.071)
Scalar
0.015
0.065
(0.063–0.068)
Strict
0.020
0.082
(0.080–0.085)
By age
Configural
0.010
0.082
-
(0.078–0.086)
Metric
0.020
0.069
Configural – Metric
(0.066–0.072)
Scalar
0.022
Strict
0.020
0.063
(0.063–0.065)
0.082
(0.080–0.085)
By educational level
Configural
0.009
0.077 (0.073–0.081)
Metric
0.011
0.061
Scalar
0.017
-
-
-
-
0.000
0.002
−0.015
Metric – Scalar
−0.003
0.006
0.002
Scalar – Strict
−0.005
0.003
0.004
Configural – Metric
(0.058–0.064)
0.063
(0.061–0.066)
Strict
0.019
0.067
(0.065–0.069)
study used traditional indices (1CFI) and alternative indices
that have been proposed in recent years (1SRMR and 1
RMSEA) to obtain additional evidence. It was therefore
possible to observe that some scalar invariance hypotheses
were rejected when more than one fit index was compared.
Likewise, we should note that the confirmation of certain
measurement invariance hypotheses does not mean there are
no variations between the attributes of the different groups
under comparison. What it means is that the instrument
is able to efficiently measure, and with less error, between
the different groups, without affecting the measurements,
which increases the internal validity of the inferences that
can be drawn. The results showed, for example, that the
hypotheses of configural invariance and metric invariance are
sustained across educational levels, whereas the scalar and
strict hypotheses are rejected. This evidence suggests that the
anxiety characteristics measured by the GAD-5 are present
at all four educational levels (configural invariance) and that
the metric for measuring anxiety in each level is identical
(metric invariance). However, the latent averages (intercepts)
obtained from the measurements between the different levels
level, and scalar invariance in the comparison by sex and
age. This provides evidence that the use of the GAD-5 as
a screening instrument in the general population allows for
adequate comparisons between men and women and between
age groups.
The results of the measures of configural, metric, and
scalar invariance, both by sex and by age group, show that
the construct (factor loadings) and the levels of the underlying
items (intercepts) are equal in all the groups tested. Accordingly,
these groups attribute the same meaning to the latent construct
studied, and their scores on the latent variable can be compared.
Although strict variance was not achieved, indicating that the
explained error variances are not equal in all groups, they can
still be compared with respect to the latent variable. It should be
noted that the latent variable is measured with different degrees
of error between groups (10). However, provided that at least
two loadings and intercepts are the same across groups, valid
inferences can be made about the differences between the means
of the latent factors in the model (10).
Since there is still a significant debate concerning the
fit indices to be used to assess parameter invariance, this
Frontiers in Psychiatry
06
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
vary significantly, as does the degree of error in the estimation
process (residuals).
At the same time, it is important to recognize that although
measures of configural, metric, scalar, and strict invariance
are enormously useful in the construction and evaluation of
psychological theories, their validity and existence in the real
world of psychological measurement and research can never
be definitively established in practice: they remain more of
an ideal (24). The challenge for researchers who allow for
partial invariance (in other words, that evidence is not obtained
for all types of invariance) is to determine how much noninvariance can be tolerated while still claiming to measure the
same construct across groups: they must make a decision based
on the anticipated threat to the validity of their findings in each
course of action (25). Novel approaches have been proposed for
the use of partial invariance analysis through simulations, and it
has been suggested that these can outperform total and partial
invariance approaches when there are many small differences in
item parameters (26).
Despite these considerations, the GAD-5 is a useful
alternative in the general population that can be used in primary
care settings, like the GAD-7 (11, 12, 27–31), and during health
emergencies such as the COVID-19 pandemic. In this respect,
the GAD-5 offers the practicality of web-based application in
addition to the novelty of the response format used. These
features contribute to the current debate on how the number of
response options affects the psychometric properties of Likerttype scales (32, 33): it has been reported that reliability increases
and excessive interpolation is avoided when response options
increase from five to seven (34, 35), a result that could be more
evident in online surveys.
Finally, it is important to consider the need to identify
anxiety-like symptomatology even if it has only been present
for a short time, and the GAD-5 refers to the previous 2 weeks.
Short periods of anxiety have been reported to be predictive
of subsequent psychopathology and may present as much
associated disability at 6-month follow-up as longer periods (3).
Including these screening options in routine care settings could
therefore be a highly effective preventive action for the detection
of common mental disorders in primary care, and improve the
level of detection and diagnosis of these disorders in public
health systems (3, 36).
more mental health issues than their cisgender peers, including
higher rates of depression, suicide, violence, and drug use
(37). By achieving parameter invariance in these groups,
we could confirm whether variations are due to the level
of anxiety presented by the person, irrespective of group
membership. There is thus a need to obtain scientific evidence
regarding this sexually diverse population to support its
mental health by strengthening the competencies of health
system professionals, and also for the formulation of public
policy (38).
We must also recognize that cross-sectional measurement
does not allow for the exploration of invariance over time,
which is also important (39). To do so, it would be necessary
to conduct follow-up measurements to assess long-term effects
in the population, which was beyond the initial scope of
the mental health strategy during the COVID-19 pandemic.
Future studies could evaluate the partial invariance of the
GAD-5 parameters at levels that could not be confirmed
in this study, for example at the educational level, and
for scalar and strict invariance in all cases. We also think
it is important to evaluate other variables of interest, but
given that our study was a secondary analysis, this was not
possible. Finally, the absence of additional validation criteria and
comparative studies of the validity and usefulness of the GAD5 could also be considered a limitation of the study requiring
future research.
Despite these limitations, our results show that the scale
performs quite satisfactorily, and this allows us to make
several observations. First, it is possible to use the scale
without the need for any special adjustment or scoring to
detect anxiety levels in the population, in contrast to other
measures that are used indiscriminately without knowing their
psychometric properties or whether they require specific scoring
to accurately place examinees on a continuum. Second, the
scale allows for comparisons between examinees, regardless
of their age, educational level, or sex, since the data show
invariance across these variables, facilitating direct comparisons
without the need for linear transformations to compare
populations. Third, the five attributes measured by the GAD5 are sufficiently general as to be present in all of the
groups compared, which in itself constitutes evidence of
external validity.
Limitations
Conclusion
Although our data represent a robust sample of the
Mexican population, it should be noted that data collection
was conducted entirely online, which may lead to participation
as well as information bias. At the same time, by considering
only the categories of male and female, we omitted transgender,
nonbinary, and gender-diverse individuals, who experience
The GAD-5 shows a unidimensional theoretical structure
and configural, metric, and scalar invariance in its comparisons
by sex and by age group, which supports its use as a screening
instrument in the general population. Since it is a short,
easily administered instrument, its use could make a crucial
contribution to the identification and treatment of mental health
Frontiers in Psychiatry
07
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
Funding
problems in both the general population and the primary care
setting. This study adds to the growing evidence about the
concise and simple GAD-7 questionnaire, demonstrating that
its five-item version, the GAD-5, could facilitate its application
in primary care settings. The brevity and predictive value of
this scale suggest its potential value as an initial assessment
tool for clinicians that facilitates timely intervention to treat
these disorders.
This work presents some results of the research project
HIM/2015/017/SSA.1207 Effects of mindfulness training on
psychological distress and quality of life of the family
caregiver, Main researcher: Filiberto Toledano-Toledano, Ph.D.
The present research was funded with federal funds for
health research and was approved by the Commissions of
Research, Ethics and Biosafety [Comisiones de Investigación,
Ética y Bioseguridad], Hospital Infantil de México Federico
Gómez, Instituto Nacional de Salud. The funding agency
had no control over the design of the study; the collection,
analysis and interpretation of the data; or the writing of
the manuscript.
Data availability statement
The data presented in this study are available on request
from the corresponding author.
Ethics statement
Conflict of interest
The study was reviewed and approved by the Research Ethics
Committee of the UNAM Psychology Faculty, registration
number FPSI/422/CEIP/157/2020. All data are protected under
the computer security standards of Mexican personal data
protection laws. The participants provided their written
informed consent to participate in this study.
The authors declare that the research was conducted in
the absence of any commercial or financial relationships
that could be construed as a potential conflict
of interest.
Publisher’s note
Author contributions
All claims expressed in this article are solely those
of the authors and do not necessarily represent those
of their affiliated organizations, or those of the publisher,
the editors and the reviewers. Any product that may be
evaluated in this article, or claim that may be made by
its manufacturer, is not guaranteed or endorsed by the
publisher.
CA-G and FA-C: conceptualization, methodology, and
formal analysis. CA-G, FA-C, LR-R, LR-S, JG-G, MS-M, AJ-T,
and FT-T: research. RR, SM-C, AL-M, and CC-R: data curation.
CA-G, FA-C, and AJ-T : writing of original draft. All authors
participated in the writing, review, editing, read and approved
the final manuscript.
References
updated meta-analysis of community-based studies. Prog Neuropsychopharmacol
Biol Psychiatry. (2021)109:110207. doi: 10.1016/j.pnpbp.2020.110207
1. GBD 2019 Mental Disorders Collaborators. Global, regional, and national
burden of 12 mental disorders in 204 countries and territories, 1990–2019: a
systematic analysis for the Global Burden of Disease Study 2019. Lancet Psychiatry.
(2022) 9:137–50. Available online at: https://linkinghub.elsevier.com/retrieve/pii/
S2215036621003953. doi: 10.1016/S2215-0366(21)00395-3
6. Shevlin M, Butter S, McBride O, Murphy J, Gibson-Miller J, Hartman
TK, et al. Measurement invariance of the Patient Health Questionnaire (PHQ9) and generalized Anxiety Disorder scale (GAD-7) across four European
countries during the COVID-19 pandemic. BMC Psychiatry. (2022) 22:1–
9. doi: 10.1186/s12888-022-03787-5
2. Stein DJ, Kazdin AE, Ruscio AM, Chiu WT, Sampson NA, Ziobrowski
HN, et al. Perceived helpfulness of treatment for generalized anxiety disorder:
a World Mental Health Surveys report. BMC Psychiatry. (2021) 21:1–14.
doi: 10.1186/s12888-021-03363-3
7. Mulvaney-Day N, Marshall T, Downey Piscopo K, Korsen N, Lynch S,
Karnell LH, et al. Screening for behavioral health conditions in primary care
settings: a systematic review of the literature. J Gen Intern Med. (2018) 33:335–
46. doi: 10.1007/s11606-017-4181-0
3. Goldberg DP, Reed GM, Robles R, Minhas F, Razzaque B, Fortes S,
et al. Screening for anxiety, depression, and anxious depression in primary
care: a field study for ICD-11 PHC. J Affect Disord. (2017) 213:199–
206. doi: 10.1016/j.jad.2017.02.025
8. Goldberg DP, Prisciandaro JJ, Williams P. The primary health care version
of ICD-11: the detection of common mental disorders in general medical
settings. Gen Hosp Psychiatry. (2012) 34:665–70. doi: 10.1016/j.genhosppsych.2012.
06.006
4. Haller H, Cramer H, Lauche R, Gass F, Dobos GJ. The prevalence and burden
of subthreshold generalized anxiety disorder: a systematic review. BMC Psychiatry.
(2014) 14:1–13. doi: 10.1186/1471-244X-14-128
9. Romano I, Ferro MA, Patte KA, Leatherdale ST. Measurement invariance of
the GAD-7 and CESD-R-10 among adolescents in Canada. J Pediatr Psychol. (2021)
47:1–10. doi: 10.1093/jpepsy/jsab119
5. Santabárbara J, Lasheras I, Lipnicki DM, Bueno-Notivol J, Pérez-Moreno
M, López-Antón R, et al. Prevalence of anxiety in the COVID-19 pandemic: an
Frontiers in Psychiatry
08
frontiersin.org
Astudillo-García et al.
10.3389/fpsyt.2022.973134
10. van de Schoot R, Lugtig P, Hox J, A. Checklist for testing
measurement invariance. Eur J Develop Psychol. (2012) 9:486–
92. doi: 10.1080/17405629.2012.686740
26. van de Schoot R, Kluytmans A, Tummers L, Lugtig P, Hox J, Muthén B.
Facing offwith Scylla and Charybdis: a comparison of scalar, partial, and the
novel possibility of approximate measurement invariance. Front Psychol. (2013)
1–15. doi: 10.3389/fpsyg.2013.00770
11. Spitzer RL, Kroenke K, Williams JBW, Löwe B, A. brief measure for assessing
generalized anxiety disorder: The GAD-7. Arch Intern Med. (2006) 166:1092–
7. doi: 10.1001/archinte.166.10.1092
27. Staples LG, Dear BF, Gandy M, Fogliati V, Fogliati R, Karin E, et al.
Psychometric properties and clinical utility of brief measures of depression,
anxiety, and general distress: the PHQ-2, GAD-2, and K-6. Gen Hosp Psychiatry.
(2019) 56:13–8. doi: 10.1016/j.genhosppsych.2018.11.003
12. Moreno E, Muñoz-Navarro R, Medrano LA, González-Blanch C, RuizRodríguez P, Limonero JT, et al. Factorial invariance of a computerized
version of the GAD-7 across various demographic groups and over time in
primary care patients. J Affect Disord. (2019) 252:114–21. doi: 10.1016/j.jad.
2019.04.032
28. García-Campayo J, Zamorano E, Ruiz MA, Pérez-Páramo M,
López-Gómez V, Rejas J. The assessment of generalized anxiety disorder:
psychometric validation of the Spanish version of the self-administered
GAD-2 scale in daily medical practice. Health Qual Life Outcomes. (2012)
10:1–10. doi: 10.1186/1477-7525-10-114
13. Teymoori A, Real R, Gorbunova A, Haghish EF, Andelic N, Wilson
L, et al. Measurement invariance of assessments of depression (PHQ-9) and
anxiety (GAD-7) across sex, strata and linguistic backgrounds in a Europeanwide sample of patients after Traumatic Brain Injury. J Affect Disord. (2020)
262:278–85. doi: 10.1016/j.jad.2019.10.035
29. Moriana JA, Jurado-González FJ, García-Torres F, Contreras A, MuñozNavarro R, González-Blanch C, et al. Exploring the structure of the GAD-7 scale
in primary care patients with emotional disorders: a network analysis approach. J
Clin Psychol. (2022) 78:283–97. doi: 10.1002/jclp.23217
14. Putnick DL, Bornstein MH. Measurement invariance conventions and
reporting: the state of the art and future directions for psychological research.
Develop Rev. (2016) 41:71–90. doi: 10.1016/j.dr.2016.06.004
30. McDonald AJ, Wickens CM, Bondy SJ, Elton-Marshall T, Wells S,
Nigatu YT, et al. Age differences in the association between loneliness and
anxiety symptoms during the COVID-19 pandemic. Psychiatry Res. (2022)
310:114446. doi: 10.1016/j.psychres.2022.114446
15. Morales Chainé S, López Montoya A, Bosch Maldonado A, Beristain Aguirre
A, Robles García R, Garibay Rubio CR, et al. Mental health symptoms, binge
drinking, and the experience of abuse during the COVID-19 lockdown in Mexico.
Front Public Health. (2021) 9:979. doi: 10.3389/fpubh.2021.656036
31. Lin S. (Lamson). Generalized anxiety disorder during COVID-19 in Canada:
gender-specific association of COVID-19 misinformation exposure, precarious
employment, and health behavior change. J Affect Disord. (2022) 302:280–
92. doi: 10.1016/j.jad.2022.01.100
16. CONAPO. Índices de desarrollo social, 2000 [Internet]. (2003). Available
online at: http://www.conapo.gob.mx/work/models/CONAPO/Resource/1342/1/
images/03Indicedesarrollo.pdf
32. Dawes J. Do data characteristics change according to the number of scale
points used? An experiment using 5-point, 7-point and 10-point scales. Int J Market
Res. (2008) 50:61–77. doi: 10.1177/147078530805000106
17. Matell MS, Jacoby J. Is there an optimal number of alternatives for likert
scale items? Study 1: Reliability and validity. Educ Psychol Meas. (1971) 31:657–
74. doi: 10.1177/001316447103100307
33. Lozano LM, García-Cueto E, Muñiz J. Effect of the number of response
categories on the reliability and validity of rating scales. Methodology. (2008)
4:73–9. doi: 10.1027/1614-2241.4.2.73
18. Alwin DF. Feeling thermometers versus 7-point scales: Which are better?
Sociol Methods Res. (1997) 25:318–40. doi: 10.1177/0049124197025003003
34. Dillman DA, Smyth JD, Christian LM. Internet, Mail and Mixed-Mode
Surveys. Hoboken, New Jersey: Wiley. (2014).
19. Rosseel Y. lavaan: An R package for structural equation modeling. J
Stat Softw. (2012) 48:1–36. Available online at: http://www.jstatsoft.org/v48/
i02/. doi: 10.18637/jss.v048.i02
35. Finstad K. Response interpolation and scale sensitivity: evidence against
5-point scales. J Usability Stud. (2010) 5:104–10.
20. Satorra A, Bentler PM, A. Scaled difference chi-square test
statistic for moment structure analysis. Psychometrika. (2001) 66:507–
14. doi: 10.1007/BF02296192
36. Mughal AY, Devadas J, Ardman E, Levis B, Levis B, Go VF, et al. A
systematic review of validated screening tools for anxiety disorders and PTSD
in low to middle income countries. BMC Psychiatry BioMed Central. (2020)
20:1–18. doi: 10.1186/s12888-020-02753-3
21. Kline RB. Principles and practice of structural equation modeling (3rd ed). New
York: The Guilford Press. (2011).
22. Cheung GW, Rensvold RB. Evaluating
for testing measurement invariance. Struct
9:233–55. doi: 10.1207/S15328007SEM0902_5
37. Newcomb ME, Hill R, Buehler K, Ryan DT, Whitton SW, Mustanski B.
High burden of mental health problems, substance use, violence, and related
psychosocial factors in transgender, non-binary, and gender diverse youth and
young adults. Arch Sex Behav. (2020) 49:645–59. doi: 10.1007/s10508-019-01533-9
goodness-of-fit indexes
Equ Model. (2002)
23. Chen FF. Sensitivity of goodness of fit indexes to lack of
measurement
invariance.
Struct
Equ
Modeling.
(2007)
14:464–
504. doi: 10.1080/10705510701301834
24. Meredith W. Measurement invariance, factor analysis and factorial
invariance. Psychometrika. (1993) 58:525–43. doi: 10.1007/BF02294825
38. Dolotina B, Turban JL, A. Multipronged, evidence-based
approach to improving mental health among transgender and
gender-diverse youth. JAMA Network Open Am Med Assoc. (2022)
5:e220926. doi: 10.1001/jamanetworkopen.2022.0926
25. Guenole N, Brown A. The consequences of ignoring measurement
invariance for path coefficients in structural equation models. Front Psychol. (2014)
5:980 doi: 10.3389/fpsyg.2014.00980
39. van de Schoot R, Schmidt P, de Beuckelaer A. Measurement invariance
[Internet]. In van de Schoot R, Schmidt P, de Beuckelaer A, editors. Frontiers Media
S.A. (2015). (Frontiers Research Topics). doi: 10.3389/978-2-88919-650-0
Frontiers in Psychiatry
09
frontiersin.org