Professional Documents
Culture Documents
Abstract
Objective: The aim of this study was to investigate the psychometric properties and diagnostic accuracy of phoneme elision
task (PET).
Method: We assessed cross-sectionally 470 Brazilian children (54.3% girls) aged between 7 and 11 years (mean age = 8.83,
sd = 0.85), from the 2nd to 4th grades. Children were assessed in their phonemic awareness ability, as well as intelligence,
general school achievement, both verbal and visuospatial working memory, single-word reading, and nonsymbolic magnitude
comparison. Beyond the psychometric properties and diagnostic accuracy of PET, we also provided reference values.
Results: Our data suggest that PET is composed mainly of one single construct, with high item reliability and precision (KR-20
above 0.90). In general, items have acceptable discriminability, considering item-total correlations. Overall PET is generally a
good screening tool for reading and spelling difficulties (SD), as well as to identify children with learning difficulties in the early
grades. However, it is not a reliable measure for screening math learning difficulties. Finally, PET shows good convergent and
divergent validity.
Conclusions: We provide evidence about the psychometric properties and diagnostic accuracy of a PET. Results contribute to
the assessment of phonemic awareness in Brazilian children, in both clinical and research contexts. The PET can be used as a
screening tool for reading and SD, which could lead to early interventions.
Keywords: Phonemic awareness; Phoneme elision; Learning skills; Spelling; Reading; Arithmetics
Introduction
Phonological processing abilities have been implicated in school learning in several ways. The term phonological processing
refers to a heterogeneous set of abilities related to accessing, discriminating, and manipulating phonological forms in short-term
memory (Wagner & Torgesen, 1987). Initially, research focused on phonological processing abilities as a correlate of visual word
decoding in alphabetic orthographies (Wagner &Torgesen, 1987). Evidence indicates that speed of access to the phonological
forms (Logan, Schatschneider, & Wagner, 2011), phonological working memory (Knoop-van Campen, Segers, & Verhoeven,
© The Author(s) 2020. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
doi:10.1093/arclin/acz085
2 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
2018), and phonemic awareness (Melby-Lervåg, Lyster, & Hulme, 2012; Hulme & Snowling, 2015) are important correlates,
and sometimes predictors, of learning to read words.
The phonemic level of representation has been implicated as crucial to visual word decoding. Processing at the phonemic
level is usually assessed with tasks of reading pronounceable nonexistent words (pseudowords) and phoneme manipulation tasks
Participants
Procedure
All study procedures complied with the Helsinki principles of research for human subjects and were approved by the local
research ethics board (COEP-UFMG). The instruments were administered by psychology undergraduate and graduate students.
The PET was developed to assess phonemic awareness in the context of reading/writing and arithmetic learning disabilities
(Lopes-Silva et al., 2014, 2016). It was associated with both reading and writing of words and numbers on these comparable
samples. Nevertheless, it was not previously analyzed in terms of its psychometric properties.
Due to the large period of data collection, changes in the assessment protocol were made. These changes included updating
some norms and replacing the reading measure to a more comprehensive one. Therefore, as will be described in detail below, we
stratified our sample according to the reading test they performed. The initial protocol used the Brazilian School Achievement
Test (Teste do Desempenho Escolar—TDE) reading subtest during 2 years. Data collected between 2014 and 2017 used the
Single-Word Reading test (Leitura de PalavrasIsoladas—LPI), which was further included in the protocol due to its nonword
items that could be more informative regarding the subtypes of reading difficulties.
Instruments
Children were assessed at their schools. The tasks were administered in two steps: group assessment of school achievement
(math and spelling) and intelligence, followed by an individual neuropsychological assessment using the subsequent tasks. First,
we describe the Phoneme Elision Test. Then, we discuss the other tasks used.
Phoneme elision task. Phoneme elision tests are a widely accepted measure of phonemic awareness, considered to be one of
the best cognitive markers of single-word reading (Wagner & Torgesen, 1987; Castles & Coltheart, 2004; Melby-Lervåg et al.,
2012). The child hears a word pronounced by the examiner and then states which word would result if a specific phoneme was
deleted. The test comprises 28 items: in eight of them, the child must delete a vowel, and in the other 20, a consonant. The
consonants to be suppressed varied by place and manner of articulation. The phoneme to be suppressed could be in different
positions within the words, which ranged from two to three syllables. After the exclusion of the phoneme, the item became
another real word. For example: in Portuguese, “atLas” without /l/ gives “atas,”“perUa” without /u/ gives “pera,” etc. Similar
examples in English would be “Farm” without /f/ giving “arm” and “Cup” without /k/ giving “up.” The PET’s reliability is
presented in the results section.
Raven’s coloured progressive matrices. The CPM is a nonverbal test used to assess fluid intelligence (Carpenter, Just, & Shell,
1990, Raven, 2000). The Brazilian validated version was used (Angelini, Alves, Custódio, Duarte, & Duarte, 1999), and the
analyses were based on Z-scores calculated from the manual’s norms (Cronbach’s α = 0.82).
Brazilian School Achievement Test (Stein, 1994; Oliveira-Ferreira et al., 2012). The TDE is the most widely used standardized
test of school achievement in Brazil. Norms are available from the 1st to 6th grades. It comprises three subtests: Mathematics
(measuring arithmetic abilities), single-word Spelling, and single word Reading. The Mathematics subtest is composed of three
simple orally presented word problems (e.g., “which number is larger, 28 or 42?”) and 35 written arithmetic calculations
of increasing complexity (e.g., easy: “4–1 = ?”; intermediate: “1230 + 150 + 1620 = ?”; and hard: “823 × 96 = ?”;
“3/4 + 2/8 = ?”). The Mathematics subtest exclusively assesses arithmetic calculation abilities. The Spelling subtest consists
4 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
of dictation of 34 single words of increasing syllabic complexity (e.g., “toca,”“Balanço,” and “cristalização”). The single-word
Reading subtest of the TDE consists of 75 single-word stimuli, which must be read aloud by the participant. Only the Reading
subtest was assessed individually. Reliability coefficients (Cronbach’s α) of TDE subtests in our samples are 0.87 or higher.
Children are instructed to work on the problems to the best of their capacity, without time limits.
Corsi blocks. This test requires that children tap wood blocks in the same order as the administrator, in forward and backward
order. The forward order is used to assess visuospatial short-term memory (Cronbach’s α = 0.62) and the backward order
to assess visuospatial working memory (Cronbach’s α = 0.69), according to the procedures systematized by Kessels, van
Zandvoort, Postma, Kappelle, and de Haan (2000). We also calculated the total score (correct trials × span) for both orders
of application.
Single-Word Reading test (Salles, Piccolo, Zamo, & Toazza, 2013). The LPI was used to assess child single-word reading
ability. In the LPI task, the child must read a list of 60 single-presented stimuli, equally distributed into regular, irregular, and
pseudowords. Stimuli are presented individually over a blank background, using black Arial 24 font in a computer screen.
Real words vary according to psycholinguistic characteristics of regularity, frequency, and length. Words shorter than five
letters are considered disyllabic and those with six–eight letters are considered polysyllabic. Stimuli were selected according
to the frequency of occurrence of words in Brazilian early school textbooks (Pinheiro, 1996). Pseudowords are constructed
combining graphemes to form letter strings with no meaning, which obey phonotactic constraints of Brazilian Portuguese.
Internal consistency in our sample was good (α = 0.88).
Nonsymbolic magnitude comparison task (Pinheiro-Chagas et al., 2014). The nonsymbolic numerical magnitude comparison
task was used to examine the divergent validity of the PET. The nonsymbolic numerical magnitude comparison task assesses
the internal Weber fraction (w), a measure of the accuracy of nonsymbolic number representations. Internal Weber fraction
(w) has been associated with math achievement but is independent from symbolic forms of processing (Pinheiro-Chagas
et al., 2014, Schneider et al., 2017). In this task, the participants are instructed to compare two simultaneously presented sets of
dots, indicating which one contains the larger number. Black dots are presented on a white circle over a black background in
a computer screen. Stimuli were designed to avoid confusion between discrete and continuous dimensions of numerosity (see
Pinheiro-Chagas et al., 2014, for details). The task comprises 8 learning trials and 64 experimental trials. Maximum stimulus
presentation time is 4,000 ms, and intertrial interval is 700 ms. Before each trial, a fixation point appears on the screen—a cross,
printed in white, with 30 mm in each line. If the child judges that the right circle presents more dots, a predefined key on the right
side of the keyboard should be pressed with the right hand. On the contrary, if the child judges that the left circle contains more
dots, then a predefined key on the left side should be pressed with the left hand (Costa et al., 2011). As a measure of approximate
number system (ANS) acuity, the internal Weber fraction (w) is calculated for each child based on the Log-Gaussian model of
number representation (Dehaene & Cohen, 2007) with the methods described by Piazza, Izard, Pinel, Le Bihan, and Dehaene
(2004).
Statistical analyses
Four aspects of PET’s psychometric properties were examined: structural validity, internal consistency, criterion validity,
and convergent/divergent validity. To assess the structural validity of the task, we ran an exploratory factor analysis (EFA)
to determine the dimensionality of PET using Factor Analysis version 10.5.03 (Lorenzo-Seva & Ferrando, 2006, 2013). To
investigate the reliability, we calculated its internal consistency using the Kuder–Richardson Formula 20 (KR-20), since the
items were coded as dichotomous variables. Analyses of structural validity and internal consistency were conducted on the
whole sample of 470 children.
Criterion validity of the PET to identify specific learning difficulties was assessed by examining its association with two
different measures of single-word Reading, one measure of word Spelling and one of Mathematics. We subdivided the samples
according to the reading test they performed. In the TDE subsample, 202 children from the 2nd to 4th grades performed the TDE
Reading, Spelling, and Mathematics subtests. In the other, LPI subsample, 199 children from the 3rd to 4th grades performed LPI,
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 5
TDE Spelling, and Mathematics subtests. To assess the diagnostic accuracy of PET, each of the subsamples was subdivided into
two achievement groups: children with performance below the 25th percentile (poor achievers) and children with performance
at or above the 25th percentile (good achievers) on the school achievement measures. The accuracy of PET in discriminating
pairwise between the groups of poor and good achievers on each dependent variable was assessed using Receiver Operating
Characteristic (ROC) analyses, utilizing MedCalc version 18.11.
Convergent and divergent validities were assessed by exploring the patterns of association between the PET scores and other
variables using Pearson’s correlations. Sixty-seven children did not complete the following tasks: digit span, Corsi blocks, and
nonsymbolic magnitude comparison. The Weber fraction exceeded the limit of discrimination (w > 0.6 and R2 < 0.2) for five
other participants, and their data were excluded from the analyses. Data from the more comprehensive cognitive assessment
were available for 171 children who performed the TDE Reading and 132 children who performed the LPI as measures of
single-word Reading. With the exception of the EFA and ROC, all other analyses were conducted using SPSS 20.0.
Results
We first assessed associations between PET and sex, age, and intelligence. Results are then presented for each phase
of the validity assessment of PET, in the following order: structural validity, internal consistency, criterion validity, and
convergent/divergent validity. Sociodemographic characteristics of each subsample are presented in Table 1.
Four hundred and seventy children (54.3% girls) aged between 7 and 11 years that answered all items from the PET were
included in the analyses of association with age and intelligence and in the assessment of the structural validity and internal
consistency.
There was no between sex difference in PET score t(468) = −0.87, p = 0.38, d = −0.08. PET had significant correlations
with intelligence (r = 0.36, p < 0.001), and small but significant correlations with age (r = 0.09, p < 0.05). In order to investigate
the effect of school grade in PET accuracy rate, we conducted an one-way between subjects ANOVA. There was a small but
significant effect of grade on PET accuracy rate, F(2, 467) = 3.209, p = 0.041, η2 = 0.014. Post hoc tests using Bonferroni’s
correction revealed significant differences in PET accuracy rate only between the 2nd and 4th grades (p < .05). Finally, there
was no significant difference of intelligence across school grades F(2, 467) = 0.144, p = 0.866, η2 = 0.00.
6 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
Table 2. Internal consistency of PET, item analysis, and factorial loading with one factor
Item number Stimulus Item-total correlation Cronbach’s alpha Phoneme to be Phoneme position Error rate Factor 1 Communality
item deleted suppressed
Structural validity
An EFA was conducted to investigate the task internal structure. At first, we tested for sampling adequacy. The Kaiser-Meyer-
Olkin (KMO) test was high (0.945) and the Bartlett’s sphericity test was significant (p < 0.001). Since data were binary (0 or 1),
we used a tetrachoric correlation matrix. As the data did not follow a normal distribution, the extraction method
applied was Minimum Rank Factor Analysis. We decided to retain one factor, based on the parallel analysis (Timmerman &
Lorenzo-Seva, 2011). The employed software also provides indices for unistructural validity assessment (Ferrando & Lorenzo–
Seva, 2017). The UniCo (Unidimensional Congruence) was 0.952 (values should be higher than 0.95), ECV (Explained Common
Variance) was 0.882 (this should be larger than 0.85), and MIREAL (Mean of Item REsidual Absolute Loadings) was 0.208
(values should be lower than 0.300). This suggests that PET can be treated as an unidimensional task. Total variance explained
by one factor was 53.5%. The latent ability measured by this factor was labeled as “Phonemic Awareness.” Factorial loadings
were acceptable, varying from 0.311 to 0.845. Table 2 shows factorial loading and communalities for each item.
Reliability
A high internal consistency was revealed by the KR-20 formula (r = 0.915), when examining the whole sample. The high
internal consistency was further confirmed by a split-half analysis (r = 0.884). Items were separated according to the error rates
in the sample, in a way that each half had the same number of items and similar difficulty level (Table 2). Mean error rate for
consonants was 0.304, and mean error rate for vowels was 0.250. Phonemes in the middle of the word had a higher mean error
rate (0.320) than phonemes in the beginning of the word (0.249). Item 16 (“cruZ” without /s/ giving “cru”) was the only item
suppressing phonemes in the end of the word, and the error rate was relatively low (0.155). In general, item-total correlations
were acceptable, varying from 0.359 to 0.633. The exceptions were Item 27 (“Apreço” without /a/ giving “preço”) and Item 16
(“cruZ” without /s/ giving “cru”) with correlations of, respectively, 0.180 and 0.201 (Table 2).
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 7
Table 3. Mean scores and percentile ranks in PET according to school grade
Mean (sd) scores for PET
Reference values
Table 3 shows PET percentiles relative to grades. As can be seen in Table 3, the rate of correct items increased along with
grade. Values for the 75th PR increased steadily at one point per grade from the 2nd to the 4th grade, approaching ceiling effect.
Even though the higher percentiles are no longer discriminative, the quantiles accurately differentiate among 2nd–4th grades.
Accuracy. Criterion validity analyses were based on data obtained from 401 children (see Table 1). The dependent variable
for the diagnosis of word reading difficulties was the performance on the TDE Reading in 202 children from the 2nd to 4th
grades and the performance on the LPI in 199 children from the 3rd to 4th grades. Math and spelling achievement were assessed
with the TDE, and both were the dependent variable for the diagnostic accuracy of math and spelling learning difficulties in 401
children from 2nd to 4th grades.
To investigate in deeper detail the influence of PET on poor school achievement, a comparative analysis between good and
poor achievers was performed. Groups were divided using the 25th percentile cutoff for each school achievement measure.
Results of between-group comparisons regarding sociodemographic, as well as intelligence and memory measures, are reported
in Table 4.
To guide our interpretations of AUC values, we relied on the lower limit of the confidence interval. Taking this into account
could be a more cautious interpretation of the ROC values. As can be seen on Table 5, PET accuracy in identifying children
with reading learning difficulties (using the TDE 25th percentile cutoff) was moderate in 2nd grade and in 3rd grade. However,
using the 0.7 cutoff criterion (Swets, 1988), in the 4th grade, PET was less effective. In general, considering all grades, accuracy
was also moderate. Using the same cutoff criteria for LPI, accuracy was moderate for 3rd grade. Furthermore, it is possible to
observe that PET was more accurate to discriminate children with difficulties in irregular word than in regular or pseudoword
reading. In summary, PET is a good task to identify children with reading difficulties, but accuracy decreases over the years.
In addition, PET’s accuracy to identify children with mathematical difficulties (MD) was also tested through ROC analysis.
In addition, PET’s accuracy to identify children with MD was also tested through ROC analysis. The overall results suggest
8 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
Table 4. Sociodemographic characteristics and cognitive performance in subgroups according to reading, spelling, and mathematical abilities, and between
groups
LPI Subsample TDE reading subsample
Variables CG (N = 170) RD (N = 29) χ 2 df p CG (N = 156) RD (N = 46) χ 2 df p
Age (years) 8.68 (0.64) 8.62 (0.67) 0.66 200.58 0.06 0.09 8.83 (1.02) 8.61 (0.77) −1.34 200.00 0.18 0.24
Intelligence (Zscore) 1.07 (0.69) 0.66 (0.82) 2.34 200.47 0.53 0.54 0.83 (0.72) 0.42 (0.76) −3.36 200.00 <0.00 0.56
Digit Span forward 31.16 (13.07) 26.93 (8.98) 0.55 200.41 0.76 0.38 31.29 (10.79) 28.70 (13.26) −1.36 200.00 0.18 0.21
Digit Span backward 13.49 (7.62) 9.24 (5.01) 3.44 200.92 <0.00 0.66 13.63 (7.32) 9.78 (4.56) −4.31 119.43 <0.00 0.63
F p η2 p F p η2 p
PET 20.87 (6.12) 13.62 (6.04) 38.72 <0.00 0.16 21.29 (6.32) 12.67 (6.19) 46.30 <0.001 0.19
TDE spelling subsample TDE mathematics subsample
Variables CG (N = 307) sd (N = 94) χ2 df p CG (N = 310) MD (N = 91) χ 2 df p
Sex (females) 54% 50% 0.41 1.00 0.52 53% 52% 0.07 1.00 0.79
Grade—2nd 25% 25% 0.00 2.00 1.00 25% 25% 0.19 2.00 0.89
3rd 38% 38% 38% 38%
4th 37% 37% 37% 37%
Age (years) 8.76 (0.86) 8.62 (0.7) −1.42 399.00 0.16 0.18 8.79 (0.83) 8.48 (0.79) −3.16 399.00 <0.00 0.38
Intelligence (Z score) 0.97 (0.72) 0.56 (0.75) −4.81 399.00 <0.00 0.56 0.93 (0.73) 0.68 (0.79) −2.81 399.00 <0.01 0.33
Digit Span forward 31.78 (12.87) 26.85 (7.42) −4.64 272.99 <0.00 0.47 31.96 (12.64) 26.07 (7.98) −4.21 399.00 <0.00 0.56
Digit Span backward 13.71 (7.55) 9.86 (4.90) −5.80 239.14 <0.00 0.61 13.42 (7.10) 10.74 (7.20) −3.16 399.00 <0.01 0.38
F p η2 p F p η2 p
PET 21.52 (5.7) 13.22 (6.86) 91.15 <0.00 0.19 20.78 (6.16) 15.46 (7.85) 24.59 <0.00 0.06
Note: PET = phoneme elision task; CG = control group; RD = reading difficulties; MD = mathematical difficulties; SD = spelling difficulties. Digit Span
forward and backward: total score; TDE = subsample that responded to the Brazilian School Achievement Test; LPI = subsample that responded to the Single-
Word Reading test.
that PET was not accurate for identifying mathematics difficulties in 2nd–4th grades. Finally, diagnostic power for Spelling
difficulties (SD) was inspected. Analyzing specificity and sensibility rates as well as the AUC, PET can be considered efficient
to identify poor achievers in Spelling.
Association with other cognitive variables. To carry out the convergent and divergent validity analysis, we conducted Pearson’s
correlations between PET and the following variables: intelligence, phonological short-term memory, verbal working memory,
visuospatial short-term memory, visuospatial working memory, academic performance, and nonsymbolic numerical accuracy
(Weber fraction). Due to missing data, a pairwise deletion was used. The total n for each correlation is presented in Table 6. PET
correlated weakly with intelligence, verbal and visuospatial short-term memory, and verbal and visuospatial working memory.
The correlations with academic performance were moderate. The correlation between PET and nonsymbolic numerical accuracy
was virtually null. All results are available in Table 6.
Discussion
We investigated the psychometric properties of a phonemic awareness task, the PET, in the assessment of school learning
difficulties for the present study. In total, we investigated 470 children from the 2nd to 4th grades. Our results indicated that
performance did not differ according to sex. Better performance was observed for older children, with a higher degree of
schooling. Regarding structural validity, an EFA indicated that PET is composed mainly of one single construct, with high
item reliability and precision (KR-20 above 0.90). In general, items have acceptable discriminability, considering item-total
correlations. We also presented reference values for PET, which are highly useful considering there are few open access phonemic
awareness tasks. Through ROC analysis, we found that PET is generally a good screening tool for reading and SD. As a single
instrument, PET is not a reliable measure for screening math learning difficulties. Our results also indicate that PET was better
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 9
Table 5. PET diagnostic power for each reading measure, math, and spelling achievement
Dependent variable Grade N AUC Std. error p AUC confidence interval (95%) Cutoff Spec. Sens.
Lower Upper
to identify children with learning difficulties in the early grades. PET also exhibited good divergent validity, since it had a
very small correlation with markers of presumably unrelated cognitive processes, such as nonsymbolic numerical magnitude
processing accuracy. In addition, PET performance was significantly associated with all cognitive and achievement measures
relevant for literacy acquisition. We discuss these results in further detail in the next sections.
Sex differences
No sex differences were observed regarding PET performance (d = 0.08). This result contrasts with a body of literature
indicating the superiority of girls in tasks related to visual decoding, spelling of words, and associated cognitive markers
(Stoet & Geary, 2013). Several reasons may explain this discrepancy. The phonological segment to be discriminated could
be implicated. Studies differ widely in the measures employed. Some studies investigate sex differences at the phonemic level
(Below, Skinner, Fearrington, & Sorrell, 2010; Chipere, 2014; Lundberg, Larsman, & Strid, 2012), and still others use composite
measures (Wilsenach & Makaure, 2018). Results of this literature are not always consistent (Moura, Mezzomo, & Cielo, 2009).
To the best of our knowledge, there are no meta-analyses of the specific role played by the phonemic level in sex differences in
reading/spelling. Studies also differ widely in scope, sample size, and sampling procedures. Most studies work with small, highly
selected samples, using a quasi-experimental approach. Our approach to sampling was different, as our study is demographically
based. We assessed kids from 10 schools. Although the sample size used is relatively large, effect sizes for sex differences are
usually small. Therefore, sample size could still be insufficient to observe reliable sex differences in the expected direction in a
demographically based sample.
We found a small but significant correlation between PET and general cognitive abilities assessed by the Raven’s CPM (r
∼ 0.30). This is consistent with current knowledge. In general, intelligence is a weak but significant predictor of acquisition
of visual word decoding skills and a stronger predictor of reading comprehension (Shatil & Share, 2003). In their predictive
study, Shatil and Share showed that intelligence played a larger role in explaining reading comprehension (around 44% of
10 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
Table 6. Correlations for convergent and divergent validity for each subsample
Task PET
Raven 0.36∗∗
variance) than single-word reading (around 5% of variance). The strength of correlations between phonological awareness and
intelligence observed in the literature varies widely (from to 0.20 to 0.50; Colé et al., 2018; Wolff, 2011). These differences
could be attributed to methodological differences such as design, measures, and sampling.
Schooling effects
Schooling experience was an important associate of PET performance. Group difference analyses indicated significant
differences between the 2nd and 4th grades. This difference cannot be attributed merely to age, as the correlation between
age and PET was virtually null (r = 0.09). The schooling effect on phoneme elision is in accordance with the hypotheses
of reciprocal causal relationships between phonological awareness and learning to read in an alphabetic system (Castles &
Coltheart, 2004; Melby-Lervåg et al., 2012). PET was accurate in diagnosing children with reading learning difficulties in the
2nd and 3rd grades, but not in the 4th grade, suggesting that its diagnostic accuracy decreases as children acquire experience
with reading in a regular orthography.
Phonological awareness typically develops very quickly as literacy instruction begins (Anthony & Francis, 2005), especially
due to a reciprocal causality effect (Castles & Coltheart, 2004). It is interesting to note that learning to read has an impact on
the performance of phonological awareness, although less so on the other subcomponents of phonological processing, namely
phonological short-term memory and lexical access (Torgesen, Wagne, & Rashotte, 1994). According to Anthony and Francis
(2005), the transparency of the orthography influences the rate of development of phonological awareness after children enter
primary school. For example, in their first year of schooling, German children develop phoneme awareness more quickly than
English children do.
Reviewing the Brazilian Portuguese spelling code, Scliar-Cabral (2003) states that it can be considered a rather transparent
orthography, even though some argue that its transparency is more pronounced in reading than in spelling (Parente, Silveira, &
Lecours, 1997). Capovilla, Dias, and Montiel (2007) analyzed the development of different phonological awareness components
in Brazilian elementary school children; and consistent with our findings, they reported that children’s performance increased
from the 1st to 2nd grades, and from the 2nd to 3rd grades; however, it did not significantly increase from the 3rd to 4th grades.
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 11
Diagnostic accuracy
It was also observed, using a series of ROC analyses, that PET was better at identifying reading learning difficulties in the
earlier grades. Sensitivity (true positive rates) in the earlier grades varied from 73% to 95%, which is acceptable (Glover &
According to Dehaene (2011, see also Arsalidou, Pawliw-Levac, Sadeghi, & Pascual-Leone, 2018, Peters & De Smedt,
2018), arithmetic and numerical processing embrace three types of magnitude representation: (a) the analogue system, related
to number sense, notion of numerical quantity, and estimation capacity; (b) the Arabic system, required in the execution of
calculations involving digits; and (c) the verbal system, associated with performance in exact calculations and problems presented
verbally. Tasks involving reading and mathematical skills that rely on the verbal system, such as counting, transcoding, retrieval
of arithmetic facts, and word problems activate largely overlapping regions in the left angular gyrus (De Smedt et al., 2010;
Simmons & Singleton, 2008, Peters & De Smedt, 2018). In previous reports, an association was detected between PET and
verbal to Arabic and vice-versa numerical transcoding tasks (Lopes-Silva et al., 2014, 2016). Thus, phonological processing
12 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
should be associated only with the MD profile involving verbal aspects, which require converting the terms into a verbal code,
processing this phonological information and retrieving a long-term memory response. It is important to underscore that the math
achievement task used, the TDE, does not include specific measures of numerical transcoding, relying heavily on calculation
abilities.
Structural validity
To assess the structural validity of the task, we conducted EFA. In the present study, we followed the guidelines proposed
by Izquierdo, Olea, and Abad (2014). We did not base our decision on principal component analysis. The number of factors to
be extracted was based on more than one procedure. PET had two items with relative low factor loadings (items 16 and 27).
However, according to Hair, Black, Babin, and Anderson (2014), the sample size needed to accept a factorial loading of 0.3 is
at least 350. Since our sample for EFA was higher, we could accept items 16 and 27 in the final model. Those items also had
the lower item-total correlation, indicating that they could be somewhat problematic. Item 16 (cruZ) is the only one requiring
a deletion at the end of the word, so the low correlation could indicate that more items with this configuration are needed. The
singularity of Item 27 (Apreço) is that it is the only item requiring a vowel to be suppressed at the beginning of the word. In
addition to that, Item 27 was the easiest item (error-rate = 0.12), so its low difficulty could explain the lower factorial loading.
On the other hand, Item 16 (cruZ) difficulty is similar to Item 23 (viOla). However, Item 23 presented a higher factorial loading
that was almost two times greater than that of Item 16. This could be due to the fact that other items also demand the deletion
of an internal vowel. Even though we found only one factor, we cannot definitely rule out that other confounding factors are
included in the final model.
Phonemic awareness tasks are complex and require other accessory skills, such as working memory and lexical retrieval.
Consequently, to acquire a “true phonemic awareness score,” many other tasks must be included and many other task features
must be controlled for (see Cunningham, Witton, Talcott, Burgess, & Shapiro, 2015 for a complete discussion about the
complexity of phonemic awareness tasks). Both split-half correlation and KR-20 indicated excellent reliability indexes. Other
studies have found that PETs seem to be consistent measures with high reliability (Lervåg, Bråten, & Hulme, 2009; Poulsen,
Nielsen, Juul, & Elbro, 2017).
It is noteworthy to highlight that PET has been overwhelmingly used in experimental and quasi-experimental investigations.
With the exception of a handful of tests available in English (Wagner, Torgesen, Rashotte, & Pearson, 2013, Gibbs & Bodman,
2014; Schrank, Mather, & McGrew, 2014), studies usually do not focus on the psychometric properties of this task. This is
especially true of the few Brazilian Portuguese phonemic awareness tasks (Godoy, Fortunato, & Paiano, 2014 and Godoy &
Cogo-Moreira, 2015 for a complete revision). Therefore, in that sense, the present study is important, since it presents a task with
rigorous psychometric analysis to investigate PET’s structural validity that can be used by Brazilian researchers and clinicians.
Our results indicate a significant association between PET and measures of short-term and working memory in the
phonological (Digit Span) and visuospatial (Corsi blocks) domains. Scores in the forward order are considered indexes of short-
term storing, and scores in the backward order are considered the indexes of executive functions. Our data clearly point to the
fact that PET demands both simple storing and more complex executive functioning resources in the phonological domain,
as well as executive functioning in the visuospatial domain. However, the correlations were higher for the working memory
tasks, indicating that the executive function component may have a more important role. This can be explained by the complex
relationships between phonological awareness and working memory (Knoop-van Campen et al., 2018). According to the degree
of automatizing, different composites of those cognitive resources may be required. Demands for executive functioning resources
are reduced as the child acquires experience with word reading (Altemeier, Abbott, & Berninger, 2008). Therefore, at the
beginning of reading development or in dyslexia cases, a task that requires the manipulation of phonemes may demand more of
the central executive, in comparison to tasks requiring only the retention of verbal codes. This could explain the high correlation
with working memory found in our results.
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 13
Our results also point to a significant correlation between PET and school achievement tasks such as TDE Reading, Spelling,
and Arithmetic. This indicates that the psychological processes evaluated by PET partially share variance with those constructs,
evaluated by each of these achievement tasks. Phonological awareness is one of the cognitive aspects most related to reading
ability, being considered a strong predictor for the development of this ability (Melby-Lervåg et al., 2012; Ziegler et al., 2010).
Final remarks
The present study provides evidence about the psychometric properties and diagnostic accuracy of PET. Results contribute
to the assessment of phonemic awareness in Brazilian children, in both clinical and research contexts. It is noteworthy that,
regardless of our efforts, some questions remain open.
An aspect that needs to be further addressed is the specific influence of phonemic awareness on numerical cognition. Which
aspects of numerical processing are more related to phonological processing? In earlier research, Lopes-Silva et al. (2016) found
an association between phonemic awareness and transcoding-related abilities. However, in the present study, we did not assess
any measure of more basic number processing, such as number reading or writing. Nevertheless, it is important to note that there
was some statistically significant shared variance between phonemic awareness and mathematical skills. Furthermore, phonemic
awareness can be thought of as a marker for very specific difficulties, such as arithmetical facts (De Smedt et al., 2010; Peters
& De Smedt, 2018) and number transcoding (Lopes-Silva et al., 2014, 2016).
However, three limitations of the study should be considered: at first, the predictive diagnostic power of phonemic awareness
on learning difficulties cannot be fully investigated in a cross-sectional design. Future studies should adopt longitudinal designs
to investigate the predictive power of phonemic awareness on reading and spelling abilities and on basic number processing
skills. Longitudinal studies, as well as controlled intervention studies, are necessary in order to investigate possible causal
mechanisms, especially due to the reciprocal relationship between phonemic awareness and reading performance (Castles &
Coltheart, 2004).
Second, the interplay between working memory and phonemic awareness must be considered. The PET suffers a large
influence from working memory (Lopes-Silva et al., 2014, 2016), and it is difficult to assess phonemic awareness in school-aged
children with a measure that does not also rely on working memory.
At last, another limitation is sample stratification, which reduces the statistical power of the analysis. Besides that, our results
are in line with major theoretical assumptions about the association between phonemic awareness and reading and spelling skills.
Nevertheless, this study is significant, considering the psychometric properties and diagnostic accuracy of the PET. This task
may contribute to the assessment of phonemic awareness, in both the clinical and research contexts. Phonemic awareness is
an important underlying correlates of learning disabilities. There are neither many tasks available nor studies investigating its
structural validity.
This study further contributes to the literature on reading and spelling as well as their cognitive correlates in a relatively
transparent orthography. This is especially relevant due to the Anglocentrism that characterizes reading research (Share, 2008).
Spelling systems other than English should be investigated in order to compare whether English-language findings can be
transposed to transparent orthographies.
Finally, this study provides evidence that phonemic awareness is more strongly associated with learning difficulties in the
earlier grades. Furthermore, our PET can be used as a screening tool for reading and SD, which could lead to early interventions.
14 Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16
Funding
This work was supported by grants from the Fundação de AmparoàPesquisa do Estado de Minas Gerais (FAPEMIG, APQ-
02755-SHA, APQ-03289-10, APQ-02953-14, APQ-03642-12). VGH is supported by a CNPq fellowship (ConselhoNacional de
Conflict of Interest
None declared.
References
Altemeier, L. E., Abbott, R. D., & Berninger, V. W. (2008). Executive functions for reading and writing in typical literacy development and dyslexia. Journal
of Clinical and Experimental Neuropsychology, 30(5), 588–606. doi: 10.1080/13803390701562818.
Angelini, A. L., Alves, I. C. B., Custódio, E. M., Duarte, W. F., &Duarte, J. L. M. (1999). Manual matrizesprogressivascoloridas de Raven: escala
especial[Manual Raven Color Progressive Matrices: Special Scale]. São Paulo: Centro Editor de Testes e PesquisasemPsicologia.
Anthony, J. L., & Francis, D. J. (2005). Development of phonological awareness. Current Directions in Psychological Science, 14(5), 255–259. doi:
10.1111/j.0963-7214.2005.00376.x.
Arsalidou, M., Pawliw-Levac, M., Sadeghi, M., & Pascual-Leone, J. (2018). Brain areas associated with numbers and calculations in children: Meta-analyses
of fMRI studies. Developmental Cognitive Neuroscience, 30, 239–250. doi: 10.1016/j.dcn.2017.08.002.
Barrouillet, P., Camos, V., Perruchet, P., & Seron, X. (2004). ADAPT: A developmental, asemantic, and procedural model for transcoding from verbal to Arabic
numerals. Psychological Review, 111(2), 368–394. doi: 10.1037/0033-295X.111.2.368.
Below, J. L., Skinner, C. H., Fearrington, J. Y., & Sorrell, C. A. (2010). Gender differences in early literacy: Analysis of kindergarten
through fifth-grade dynamic indicators of basic early literacy skills probes. School Psychology Review, 39(2), 240–257. Retrieved from.
https://www.researchgate.net/profile/Christopher_Skinner/publication/282367155_Gender_Differences_in_Early_Literacy_Analysis_of_Kindergarten_
through_Fifth-Grade_Dynamic_Indicators_of_Basic_Early_Literacy_Skills_Probes/links/560eecfa08ae0fc513eeb34d/Gender.
Bergmann, J., & Wimmer, H. (2008). A dual-route perspective on poor reading in a regular orthography: Evidence from phonological and orthographic lexical
decisions. Cognitive Neuropsychology, 25(5), 653–676. doi: 10.1080/02643290802221404.
Butterworth, B. (2005). The development of arithmetical abilities. Journal of Child Psychology and Psychiatry, 46(1), 3–18. doi:
10.1111/j.1469-7610.2004.00374.x.
Capovilla, A. G. S., Dias, N. M., & Montiel, J. M. (2007). Desenvolvimento dos componentes da consciênciafonológica no ensino fundamental e correlação
com nota escolar. Psico-USF, 12(1), 55–64.
Carpenter, P. A., Just, M. A., & Shell, P. (1990). What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices
Test. Psychological Review, 97(3), 404–431. doi: 10.1037/0033-295X.97.3.404.
Castles, A., & Coltheart, M. (2004). Is there a causal link from phonological awareness to success in learning to read? Cognition, 91(1), 77–111. doi:
10.1016/S0010-0277(03)00164-1.
Chipere, N. (2014). Sex differences in phonological awareness and reading ability. Language Awareness, 23(3), 275–289. doi: 10.1080/09658416.2013.774007.
Colé, P., Cavalli, E., Duncan, L. G., Theurel, A., Gentaz, E., Sprenger-Charolles, L. et al. (2018). What is the influence of morphological knowledge in the early
stages of reading acquisition among low SES children?A Graphical Modeling Approach. Frontiers in Psychology, 9, 547. doi: 10.3389/fpsyg.2018.00547.
Costa, A. J., Silva, J. B. L., Pinheiro-Chagas, P., Krinzinger, H., Lonnemann, J., Willmes, K. et al. (2011). A hand full of numbers: A role for offloading in
arithmetics learning? Frontiers in Psychology, 2, 368. doi: 10.3389/fpsyg.2011.00368.
Cunningham, A. J., Witton, C., Talcott, J. B., Burgess, A. P., & Shapiro, L. R. (2015). Deconstructing phonological tasks: The contribution of stimulus and
response type to the prediction of early decoding skills. Cognition, 143, 178–186. doi: 10.1016/J.COGNITION.2015.06.013.
De Smedt, B., Taylor, J., Archibald, L., & Ansari, D. (2010). How is phonological processing related to individual differences in children’s arithmetic skills?
Developmental Science, 13(3), 508–520. doi: 10.1111/j.1467-7687.2009.00897.x.
Dehaene, S. (1992). Varieties of numerical abilities. Cognition, 44(1–2), 1–42. doi: 10.1016/0010-0277(92)90049-N.
Dehaene, S. (2011).The Number Sense: How the Mind Creates Mathematics, Revised andUpdated Edition.New York: Oxford University Press. Retrieved
fromhttps://psycnet.apa.org/record/2011-10610-000
Dehaene, S., & Cohen, L. (2007). Cultural recycling of cortical maps. Neuron, 56(2), 384–398. doi: 10.1016/J.NEURON.2007.10.004.
Ferrando, P. J., & Lorenzo-Seva, U. (2017). Program FACTOR at 10: Origins,development and future directions. Psicothema, 29(2), 236–241. doi:
10.7334/psicothema2016.304.
Figueiredo, V. D. (2002).WISC-III: Escala de Inteligência Wechsler paraCrianças-adaptaçãobrasileira da 3a edição[WISC-III: Wechsler Intelligence Scale for
Children - 3rd Edition Brazilian Adaptation]. São Paulo: Casa do Psicólogo.
Figueiredo, V. L. M. d., &Nascimento, E. d. (2007).Desempenhosnasduastarefas do subtestedígitos do WISC-III e do WAIS-III[Performing on both WISC-III
and WAIS-III digit subtest tasks]. Psicologia: Teoria e Pesquisa, 23(3), 313–318. doi: 10.1590/S0102-37722007000300010.
Furnes, B., & Samuelsson, S. (2010). Predicting reading and spelling difficulties in transparent and opaque orthographies: A comparison between scandinavian
and US/Australian children. Dyslexia, 16(2), 119–142. https://doi.org/10.1002/dys.401.
Gibbs, S., & Bodman, S. (2014). Phonological Assessment Battery (PhAB2 Primary) (2ndPrimar). UK: GL Assessment.
Pereira et al. / Archives of Clinical Neuropsychology XX (XXXX) XXX–XXX 00 (2020); 1–16 15
Glover, T. A., & Albers, C. A. (2007). Considerations for evaluating universal screening assessments. Journal of School Psychology, 45(2), 117–135. doi:
10.1016/J.JSP.2006.05.005.
Godoy, D. M. A., & Cogo-Moreira, H. (2015). Evidences of factorial structure and precision of phonemic awareness tasks (TCFe). Paidéia (RibeirãoPreto),
25(62), 363–372. doi: 10.1590/1982-43272562201510.
Plaza, M., & Cohen, H. (2004). Predictive influence of phonological processing, morphological/syntactic skill, and naming speed on spelling performance.
Brain and Cognition, 55(2), 368–373. doi: 10.1016/J.BANDC.2004.02.076.
Poulsen, M., Nielsen, A. M. V., Juul, H., & Elbro, C. (2017). Early identification of reading difficulties: Ascreening strategy that adjusts the sensitivity to the
level of prediction accuracy. Dyslexia, 23(3), 251–267. doi: 10.1002/dys.1560.