The study originally involved a total of 344 pre-elementary children at age 5 to 6 years. The assessment was conducted in two public and one private preschools, and two public elementary schools. None of the preschools included formal literacy instruction in their curricula. Of the original 344 children, 53 were either absent on the day of the first-grade assessment or had transferred to a different school, and thus were excluded from the final sample (15.4% drop-outs).

The final sample consisted of the remaining 291 children (154 boys). Their mean age was 70 months (SD = 3.70, range 63–79) and 81 months (SD = 3.42, range 75–87) on the days of the pre-elementary and first-grade assessments, respectively. Their mean age at elementary-school entrance was 78 months (SD = 3.42, range 72–83), which was 3 to 4 months prior to the first-grade assessment. No exclusion criterion was set so as to include children with a wide range of abilities. According to the Declaration of Helsinki, assent and informed consent were obtained from all participants and their parents or legal guardians, respectively, prior to their participation in the study. Before giving their consent and assent, they were given a full and appropriate explanation of the study protocol, which had been approved by a local ethics committee (#2020C006).

The participants were assessed once in their pre-elementary year and once in their first-grade year. The Japanese school year starts in April and ends in March. All children were assessed individually by trained researchers in a quiet room at the schools they attended.

Table 2 shows descriptive statistics of Hiragana-naming and oral reading fluency measures. Pre-elementary children could name an average of 34 of 45 Hiragana Sei-on. There was a strong rightward distribution where most of the children showed near-perfect or perfect mastery (Figure 1; skewness = −1.18). Oral reading in the first grade was overall slow, since the mean z-values of reading time were near or above +2, which indicates significant prolongation (see Materials and Methods), in all four tasks. Consequently, nearly half of the children (45.4%) met the criterion for a deficient reader (Table 2). First-grade indices were negatively correlated with pre-elementary letter-naming performance (Table 2, all Holm-adjusted ps < 0.0001), which indicated that a child who could name more letters in the pre-elementary year tended to be faster in oral reading a year later.

Figure 2 shows changes in the results of binomial logistic regression (Figure 2A) and ROC (Figures 2B,C) analyses for the possible cut-offs for the Hiragana-naming benchmark. Of all the possible cut-offs (1 to 45 letters), a benchmark of 1 to 9 letters resulted in unreasonably large and unstable odds ratios in the binomial logistic regression; thus, these results are not shown in Figure 2 and were excluded from further consideration.

The results of the regression analysis (Figure 2A) showed that the adjusted odds ratio of Hiragana-naming performance remained stable at around 15, except for the lower and higher ends. This result indicated that children who failed to meet a given benchmark were far more likely to be deficient readers in first grade. Regarding the precision of regression models, AIC reached a minimum when the benchmark was set at 39 letters (311.97) and increased only slightly at 40 letters (312.01, Figure 2A). Thus, the binomial regression analysis indicated that 39 and 40 letters were good candidates for an optimal benchmark. On the other hand, the ROC analysis indicated that 44 letters was an optimal cut-off, at which the distance between the ROC curve and a diagonal reference line became maximum (i.e., Youden Index; Figures 2B,C). The prediction accuracy of Hiragana-naming performance with respect to first-grade reading fluency was overall good, with an AUC of 0.817 (95% confidence interval: 0.768–0.867).

Taken together, these results show that, although the ROC result indicated 44 letters, setting a benchmark above 40 letters was considered to be not optimal, due to a decrease in the prediction precision. More specifically, the binomial logistic regression analysis involved an apparent increase in AIC and a drop in the odds ratio, which was associated with an increase in the number of false positives as indicated by the ROC curve (Figure 2C). Therefore, we sought a benchmark of (1) 40 letters of less, where (2) the AIC of the regression model reaches a minimum or close to a minimum, and (3) the distance between the ROC curve and a diagonal reference line reaches a maximum. As a result, all three of these criteria were met when we set a benchmark at 40 letters (Figure 2). Therefore, in the present study, 40 or more of 45 Hiragana Sei-on letters was considered to be a sufficient level of letter-naming for Japanese pre-elementary children.

Table 3 shows the final binomial logistic regression model with the selected benchmark. A model fit was indicated to be appropriate with non-significant χ² in the Hosmer-Lemeshow goodness of fit test (χ² = 10.15, df = 8, p = 0.25) and a sufficiently large estimate of determination coefficient (McFadden’s pseudo R² = 0.242; McFadden, 1997). The adjusted odds ratio indicated that a child who failed to meet the pre-elementary benchmark was at high risk of becoming a deficient reader in the first grade, while the age in first grade and gender were not significantly associated with the first-grade reading status (Table 3).

Lastly, Table 4 shows the final contingency table of pre-elementary Hiragana-naming performance and first-grade reading fluency. Diagnostic efficiency indices based on Table 4 were moderate sensitivity (0.614) and negative predictive value (0.738), high specificity (0.906) and positive predictive value (0.844), and an overall correct classification rate of 0.773. As shown in Table 4, while some deficient first graders had passed the pre-elementary benchmark (i.e., false negatives; 51/291 = 17.5% of the final sample), children who failed to meet the benchmark rarely became normal readers (i.e., false positives; 15/291 = 5.15% of the final sample). Therefore, the present benchmark was considered to more strongly predict a risk, rather than a successful outcome, with respect to reading ability in the first grade.

As expected, the present benchmark required the acquisition of nearly 90% of Hiragana Sei-on letters (40/45 letters = 88.9%). In this study, 67.0% of the Japanese pre-elementary children without formal literacy instruction met this criterion (see Table 4). Although direct comparison of this result with previously reported benchmarks in an alphabet is difficult, due to a difference in age (52 vs. 70 months in the present study; Piasta et al., 2012), the average letter-name knowledge reported in similar-aged children may serve as a reference. For example, kindergarteners in Canada (English-speaking) showed mean letter-naming accuracies of 89.1% for uppercase and 74.9% for lowercase letters, and half of them scored 80.8% (21 of 26 letters) or less for lowercase letters (Evans et al., 2006). In grande-section preschoolers in France (French-speaking), the mean accuracy for uppercase letters has been reported to be 75.0% (19.5 of 26 letters; Bouchière et al., 2010). Since early education in these countries typically involves formal alphabet instruction at this age, the imperfect acquisition suggests that learning letter-names in alphabet is generally more demanding than learning Hiragana. More strikingly, Greek kindergarteners without literacy instruction showed even lower letter-name knowledge for uppercase letters (8.47 of 24 letters, 35.3% at the end of a school year), although Greek is more transparent than English or French (Manolitsis and Tafa, 2011). Accordingly, optimal benchmarks are expected to be lower for an alphabet at this age (5 to 6 years), especially for lowercase letters.

The high level of the Hiragana-naming benchmark should be due, at least partially, to the simplicity of the letter-knowledge, as mentioned earlier. Moreover, since Hiragana words can be read by the simple addition of letter-names, reading experience after the initial acquisition can reciprocally strengthen the letter-naming ability. In many Japanese children, a higher chance of mastering letter-names without formal instruction should owe much to these characteristics. In contrast, the alphabet comprises more complex letter knowledge, and a child’s natural exposure to print (e.g., book reading) does not always or necessarily include letter names. This may account for the possibility that the acquisition of alphabet-names takes more time and relies more heavily on formal instruction, so that a smaller proportion of children can reach a mastery level at the same age. These notions suggest that the nature of letter knowledge greatly affects the course and level of letter-name acquisition in a given script, which should be further clarified by taking the difference in the education systems into consideration.

Based on the present Hiragana-naming benchmark, a high proportion (over 80%) of children who failed to meet it became deficient readers, which is in line with previous findings that poor preschool alphabet-knowledge strongly predicts later reading difficulty (Gallagher et al., 2000; Catts et al., 2001; Torppa et al., 2006; Puolakanaho et al., 2007). On the other hand, meeting the benchmark did not strongly predict normal reading fluency, as opposed to the alphabet-naming benchmarks in younger English-speaking children that predominantly predicted successful outcomes (Piasta et al., 2012). Theoretically, the acquisition of Hiragana-names immediately enables children to read words and to increase fluency, so that mastering them by a certain point in the pre-elementary year could strongly predict sufficient first-grade reading fluency; however, the present result did not fully support this hypothesis.

There are two possible accounts for this result. The first is the well-known dissociation between reading accuracy and fluency in transparent scripts. While children with reading difficulties in regular orthography can attain a normal level of reading accuracy, poor reading fluency tends to persist (e.g., Katzir et al., 2004; Lallier et al., 2014; Diamanti et al., 2018), even into adulthood (Oren and Breznitz, 2005; Re et al., 2011). Given the extremely transparent nature of Hiragana Sei-on, it is highly possible that at least some of the children with false-negative results (see Tables 1, 4) had a pure fluency deficit, which had little impact on the acquisition of letter-names. Secondly, the first-grade reading assessment in the present study contained so-called “special moras” that require additional orthographic knowledge in Hiragana, such as voiced and semi-voiced letters (Daku-on and Han-daku-on) and two-letter syllables (You-on; Welty et al., 2014; Inoue et al., 2017). In general, scripts with more complex orthography are associated with a higher prevalence of developmental dyslexia (Habib and Giraud, 2013; Peterson and Pennington, 2015; Daniels and Share, 2018), which is also the case for multiple Japanese scripts (Hiragana: 0.2–1.6% vs. Kanji: 6.1–6.9%; Uno et al., 2009). Therefore, it is possible that a child can acquire Hiragana Sei-on names normally but then struggles with “special moras” and later meets the criterion of deficient reader.

A consideration of these additional requirements in Hiragana reading leads to at least two types of developmental trajectories that are associated with the prediction characteristic of the present benchmark. First, because Sei-on letter-naming is fairly easy, even within Hiragana, children who struggle at this level will continue to be challenged by fluency acquisition and adaptation to exceptional rules. These children rarely catch up to their grade level, as reflected by the small number of false positives (see Tables 1, 4), so that the benchmark could identify at-risk children with high precision. Conversely, children with potentially normal reading ability rarely fail at this level, as reflected by the high specificity. Second, children with milder reading deficit may have little problem with Sei-on naming but start to struggle when additional requirements are introduced. It is difficult for the present benchmark to detect these children; thus, meeting the benchmark does not necessarily assure a successful reading outcome, as reflected by the lower sensitivity and negative predictive value. Given this second trajectory, the ability of present pre-elementary Hiragana-naming benchmark to predict reading outcome beyond first grade, including kanji acquisition, is a topic for future studies.

Regarding alphabet-naming, while the utility of a similar benchmark will need to be determined in future studies, differences in letter-naming with Hiragana may provide some clue. Given that the acquisition of alphabet-naming is presumably more time-consuming and demanding (Evans et al., 2006; Bouchière et al., 2010), it might be difficult for the benchmark to precisely screen for high-risk children at age 5 to 6 years. In fact, a previous study reported a high precision of predicting success, but not risk, throughout the possible range of cut-offs (i.e., 1 to 26 letters) in younger preschoolers (Piasta et al., 2012). As a speculation, the attainment of certain alphabet-naming knowledge by a given age, especially without formal instruction, might reflect a child’s potential advantage in reading acquisition, such as higher cognitive abilities (Evans et al., 2006), and thus could predict a successful outcome more reliably. While it is well-established that early alphabet-name knowledge correlates with later reading abilities (Foulin, 2005; National Early Literacy Panel, 2008; Hjetland et al., 2017), the identification of appropriate benchmark-settings for this skill will require further studies that carefully consider the nature and roles of letter-names in reading.

The present study has several limitations. First, since the present longitudinal data were obtained from a limited number of children at few schools, the benchmark and its ability to predict reading fluency should be verified in larger and different samples of children in Japan to increase a practical value. Second, we only measured the speed of oral reading in the first grade as an outcome measure. Although oral reading fluency is one of the fundamental reading abilities, especially in younger readers and/or transparent scripts (Florit and Cain, 2011; Park et al., 2015), associations between pre-school/pre-elementary letter knowledge and higher-order reading abilities such as comprehension have also been reported (National Early Literacy Panel, 2008; Piasta et al., 2012). Since the ultimate goal of reading is to comprehend the text, the significance of the present benchmark should also be examined from this perspective. Third, since the present study did not include control variables, taking potentially confounding measures (e.g., general cognitive abilities) into account should be important to verify the significance of the benchmark.

Fourth, a substantial influence of environmental and genetic factors, such socio-economic status (SES), home literacy environment, and familial background, on early reading development has been reported, even in transparent scripts (Strang and Piasta, 2016; Lohvansuu et al., 2021). Thus, lack of these measures also limits our conclusions, although letter-naming in Hiragana could be less affected by environmental factors (Inomata et al., 2016). While an interest and engagement in reading might influence Hiragana-naming acquisition to a lesser extent, due to its simplicity, their influence on further literacy development, such as reading fluency, must be examined further. Finally, variations in the level of benchmarks and the significance of early letter-name knowledge among writing systems require similar types of studies, particularly in transparent alphabetic scripts (e.g., Spanish, Italian, Finnish, Korean Hangeul) and those with unique letter-name knowledge (e.g., Hebrew, Levin et al., 2002).

Nonetheless, the present findings and presumable differences between Hiragana and alphabet scripts have important educational implications. First, an appropriate goal of the letter-naming ability and relevant educational policies, such as the starting age and the necessary amount of instruction, should vary greatly among scripts. Thus, each country and region must develop empirically-defined standards, assessments, and curricula for this foundational skill of reading in preschool to early elementary children. Second, the prediction characteristics of a letter-naming benchmark tells educators what should be a focus of monitoring at different stages. For example, Japanese preschool/pre-elementary teachers should pay close attention to and immediately start interventions with children who are, or who may be, failing to meet the Hiragana-naming benchmark. At the same time, teachers of early elementary grades must be aware of the later-onset reading deficits in children who have passed the pre-elementary benchmark. On the other hand, for an alphabet script, careful growth monitoring through formal instruction may be more suitable for identifying at-risk children on the basis of letter-naming ability. Further studies on these issues can be expected to promote various aspects of early reading acquisition and education across languages, countries, and regions.