The Triple Task (TT) was developed to assess the timing and cognitive load of various writing processes, i.e., to assess their dynamics. It was originally developed by Kellogg in 1987. According to the assumptions of Hayes and Flower’s (1980) Cognitive Process Theory of Writing, the act of writing text involves certain thinking or (meta)cognitive processes, such as planning, translating, and revising, which the writer performs in a non-linear fashion during writing process. TT is therefore used to measure the dynamics of such processes. In TT, the participant must perform three tasks more or less concurrently. In the first task, the focus is on text composing. In the second task, the participant must respond as quickly as possible to simple audio stimuli that occur in a variable interval schedule. The recorded reaction time, from which the baseline reaction time is subtracted, serves as an index of cognitive effort with respect to the first task. More precisely, this task assumes that the focus on text composing (first task) and the requirement to respond quickly to audio stimuli (second task) compete for limited attentional or working memory resources. Thus, a faster reaction time implies less cognitive effort for the ongoing cognitive process (Olive et al., 2002). In the third task, right after the second task, the participant reports on the (meta)cognitive process going on at the time of the response. The TT must be trained prior to recording because the procedure itself has to be learned.

In reviewing the literature on cognitive processes in writing, it is noticeable that not many studies use TT. Consequently, there are not many studies that have used TT in the last decade (e.g., Fidalgo et al., 2014; Limpo, 2018; Limpo and Alves, 2018), yet they have made important contribution to understanding the cognitive background of writing. There are some reasonable explanations for this “unpopularity” of TT, such as: it is a technique that is quite complex and sensitive for application; it requires some additional time for training; it is not suitable for group testing; it requires the use of custom made computer settings or software; there is no standardized response box or similar input device for use with TT, so researchers use different response techniques, for example, by voice or by clicking a mouse, response box, or keyboard key with the non-dominant hand. There is also a growing popularity of keystroke logging tools (KSL) that measure writing processes while avoiding almost all the requirements of TT. Finally, it is also important to note that the community of researchers interested in the (cognitive) processes of writing is quite small but is likely to grow thanks to KSL tools. Despite its demanding application, TT arguably provides valuable insights into the dynamics of cognitive processes in writing and thus deserves a place in studies on cognitive processes in writing. Investigating the cognitive processes involved in composing texts using TT in a population with, for example, language disorders can make an important contribution to understanding the disorder, but it can also test the potential of TT for clinical purposes. Prior to this, TT must be evaluated and validated, especially if it is being used for the first time in a population with writing difficulties.

Writing is considered one of the most complex cognitive activities, being cognitively and emotionally demanding on its own right (Kellogg, 1987). Therefore, additional tasks concurrent with writing could be reactive and disrupt the underlying cognitive processes of writing, reducing the reliability and validity of such a multiple-task method. Reactivity refers to the risk that a particular action or procedure might alter the cognitive process underlying the task (Piolat et al., 2001) or that, in a particular case, the triple task performance may disrupt or misrepresent the cognitive processes underlying the writing process and written product (Olive et al., 2002). Previous research predominantly has shown that there is no interference between the triple task (or other similar tasks such as the dual-task and the think-aloud protocols) and text quality in adults (Kellogg, 1987; Penningroth and Rosenberg, 1995; Ransdell, 1995; Piolat et al., 1996). Piolat et al. (1996) found no significant effects of coupling writing and simple reaction time task (dual-task) neither on productivity (number of words), syntactic complexity (number of words per sentence), nor on some processes: fluency, and number of revisions observed in text composing. Additionally, they did not find significant differences in the cognitive effort either on the TT or on the dual-task. Penningroth and Rosenberg (1995) compared stories written with the TT (including think-aloud protocols) and without the TT and found no differences in coherence or text quality. Finally, Piolat et al. (2001) briefly reviewed literature on the measurement reactivity and retrospective report validity in the triple task method and concluded that the TT is a stable and powerful way to measure the dynamics of written composition, it provides valid insights into writing processes and does not compromise processes or text quality. However, Janssen et al. (1996) found that think-aloud protocols alone may affect complex writing tasks more than simple ones by increasing pause length within and between sentences, and between paragraphs. Similarly, Wengelin et al. (2014) combined the keystroke logging tool (ScriptLog) and TT to analyse the reactivity of TT to writing process and product variables. The study included a sample of 40 university students who wrote expository texts under two conditions - with and without TT. The data showed no effect on product characteristics such as text length (number of words and number of characters), mean word length (number of letters per word), lexical diversity or density, or keystroke logging data such as productivity (words/min), pause frequency, pause durations and proportion of revisions. While in the condition without TT, more keystrokes and more revisions were made. More revisions, to some extent, imply also more interruptions of episodes of language production. It would be valuable to observe also whether the TT condition alters the frequency or duration of bursts [R-or P-bursts, i.e., an episode of language production terminated by a revision (R) or pause (P)]. More specifically, the “beep” sound that signals the requirement to switch the attention on a task other than writing occasionally and unnaturally interrupts the writing process, which could increase the frequency and decrease the average duration of bursts.

Special care should be taken when applying such a complex method for assessing writing to a population that struggle with reading and writing, such as people with dyslexia do. While dyslexia has traditionally been defined as a discrepancy between an average or higher general cognitive ability and low reading performance, more recent understandings of dyslexia suggest that it encompasses developmental language difficulties that go well beyond reading difficulties, including difficulties with phonological processing, word use, spelling, and writing that persist into adulthood (e.g., Lefly and Pennington, 1991; Shaywitz et al., 1992; Bruck, 1993; Ramus, 2003; Ramus et al., 2003; Smith-Spark et al., 2003; Miles and Miles, 2004; Connelly et al., 2006; Moojen et al., 2020). Moreover, Berninger et al. (2006a) demonstrated that students often manage to compensate for their reading difficulties following dyslexia language treatment, so much so that they no longer appear to have a serious problem with reading, while writing difficulties persist. In the International Classification of Diseases 11th – ICD 11 (WHO, 2019/2021), dyslexia is classified as Developmental learning disorder with impairment in reading (6A03.0), with significant and persistent difficulties manifesting in learning academic skills related to reading.

Studies dealing with language disabilities and writing are still very product-oriented and more interested in the developmental perspective, with children being the participants rather than high education students or adults. To date, there have been very few process-oriented studies of the writing of individuals with dyslexia. Studies of writing by adults with dyslexia have shown that: students with dyslexia made more morphosyntactic spelling errors, memory-related errors, punctuation errors, and capitalisation errors than the control group and that the quality of their writing was poorer based on text structure (Tops et al., 2013); adults with dyslexia wrote more slowly, produced shorter essays, used more monosyllabic and fewer polysyllabic words (possibly indicating limited word choice), and produced text with more spelling errors (Sterling et al., 1998); students with dyslexia generally produced poorer texts, as measured by a holistic essay quality assessment, and produced more spelling errors; in addition, students with dyslexia showed poorer performance on ‘lower order skills’ such as spelling and handwriting fluency (Connelly et al., 2006). Torrance et al. (2016) showed in their process- and product-oriented study that high school students with dyslexia produced four times more spelling errors than the control group and had poorer text quality at the lexical, structural, and thematic levels; they also had longer inter-key-press latencies at all text locations at the pre-word, within-word, and word levels. Wengelin (2007) showed in her KSL study that adults with dyslexia paused more frequently during writing, especially at the within-word level, and made a higher proportion of spelling-related revisions; they also demonstrated a narrower vocabulary, as measured by the lexical density and lexical diversity of the texts.

Therefore, for adults with dyslexia, writing itself is more challenging and the TT might be more reactive, i.e., more disruptive to writing processes than in adults with typically developed reading and writing skills; and not just because of reading and writing difficulties, but also abilities and/or skills underlying the mentioned language skills. Coordinating multiple processes when composing texts, i.e., focusing and shifting attention between (meta)cognitive processes and tasks, and making decisions when selecting alternatives are examples of executive functions that operate in the background of thinking, language, and memory processes during text composition (Baddeley, 1996; Piolat et al., 2001). Readers with learning difficulties tend to perform lower on measures of the central executive system of working memory, indicating a lower ability to focus attention in a temporally flexible manner and to switch attention as needed (Swanson, 1999). The inability to shift attentional focus may lead to repetition of errors or persistence in repeating the same response (Lezak, 1995). In addition, studies confirm that individuals with dyslexia typically perform lower in organising, automatising, and integrating multiple processes, and perform lower on inhibition and shifting/cognitive flexibility tasks (e.g., Helland and Asbjørnsen, 2000; Berninger et al., 2006a,b; Altemeier et al., 2008). For example, there is already a large body of research showing that individuals with dyslexia have difficulty automatising reading skills but also other language related skills such as language fluency including naming speed and verbal retrieval (e.g., Denckla and Rudel, 1976; Yap and van der Leij, 1994; Berninger, 2001; Vellutino et al., 2004). The degree of automatization of a given skill has been tested by introducing a second task to be performed more or less concurrently with the first task. The performance on the dual-task is better if the individual is better able to automatise the second task. In their study, Hecht et al. (2004) compared performance in a single-task condition and a dual-task condition to test whether faster readers were more able to automatise a single cognitive task, which should subsequently help them perform better in the dual-task situation. The results showed that faster adult readers automatised the predictable cognitive task more quickly and performed better on the dual-task than slower readers. They suggested that slower readers need more practice to achieve optimal performance.

Such a dual-task method is widely used in cognitive psychology (e.g., Kellogg, 1994; Baddeley and Andrade, 2000), whereas the TT may be even more challenging for a fluent performance because it contains also a third requirement. Moreover, the second task is hardly predictable because the audio signal occurs in a variable interval schedule, while the third task is more predictable because it has only three options, i.e., three choices of metacognitive processes to be reported introspectively (planning, translating, or revising). Unpredictable and varied, unlike predictable and constant tasks, are harder to automatise (Shiffrin and Dumais, 1981; Hecht et al., 2004), which provides an opportunity to measure cognitive effort in the secondary task of TT. On the other hand, if the audio signals are predictable, i.e., if they occur in a certain rhythm, then reaction time does not represent cognitive effort, because participants can anticipate the audio signal and respond automatically (or rhythmically), probably even without paying attention to the audio signal. However, TT involves three tasks that must be performed fluently and (more or less) concurrently by flexibly switching focus between tasks and keeping maintaining attention on the writing task. TT performance in individuals with typical reading skills is expected to become more fluent and, to some extent, also more automatic with training, i.e., without difficulty in performing three tasks in the correct order (focusing on the first task and, after audio stimuli, performing the second and then third tasks), in learning the instructions for direct introspection of metacognitive processes in the third task, or in directing focus to the first task and not to the other two. Perhaps it is more challenging for adults with dyslexia to produce such a fluent performance than for those with typical reading skills, which could manifest as reactivity of TT in the dyslexia group. While previous studies have shown that such trained TT performance has no effect on the writing process and writing product of individuals with typical reading skills, at least not with known measures of writing process and writing product (e.g., Penningroth and Rosenberg, 1995; Ransdell, 1995; Piolat et al., 1996), it is not known if the same is true for individuals with dyslexia. There is a reasonable concern that adults with dyslexia may have difficulty performing TT fluently, which could be reflected in writing processes or text quality characteristics. Therefore, it is important to measure possible reactivity effects when TT is used with individuals with dyslexia.

To sum up, TT reactivity to writing processes or product has not been demonstrated in most previous studies with populations with typical language skills (e.g., Penningroth and Rosenberg, 1995; Piolat et al., 1996), with the exception of Wengelin et al. (2014), who found more keystrokes and more revisions in the condition without TT. However, TT reactivity has not yet been tested in populations with language disorders, such as individuals with dyslexia. There is also a lack of data obtained by combining TT with KSL tools to measure TT reactivity.

Therefore, the first goal of this study was to investigate whether the TT method affects writing processes and/or written product in adults with dyslexia compared to a control group of adults with typical reading skills. It might be expected that adults with dyslexia may have difficulty in performing concurrently three tasks fluently, which could be reflected on some writing processes characteristics or in the text quality. Although many studies have not found evidence of TT reactivity in typical readers, some have, and studies with readers with dyslexia are non-existent. The second goal was to investigate whether the TT method affects the writing process and/or written product independently of dyslexia. From previous studies, it appears that some process measures could be disrupted by TT, e.g., revisions. And the third goal is to investigate group differences, i.e., to compare adults with dyslexia and adults with typical reading skills in terms of text quality and writing process measures.

To examine whether the triple task had a differential effect on text quality between groups of adults with dyslexia and TRS adults, mixed ANOVAs were conducted, using IBM SPSS Statistics 26 (IBM Corp, 2019). Group membership was a between-subjects factor, whereas measures of writing processes and text quality for two different texts (one with and one without TT) were within-subjects’ factors. The interaction effects of group (dyslexia vs. TRS) and TT condition (with vs. without) were tested, as well as the main effects of TT condition and group on measures of writing processes and text quality. The first main effect shows differences between two TT conditions independent of group, the second differences between two groups, independent of TT condition, in text quality and writing process variables.

Comparing to frequentists approach, Bayes statistics provide a coherent method to determining whether non-significant results support a null hypothesis over a theory, or whether the data are just insensitive (Dienes, 2014). Frequentist and Bayesian methods have different assumptions about the data and model parameters. Frequentist statistics assume that the null hypothesis is true and calculate the probability of obtaining the observed data, whereas Bayesian statistics incorporate prior assumptions and update them based on the observed data, providing evidence either in favour of or against the null hypothesis. While p-values in frequentist analyses can only reject the null hypothesis, the Bayes factor can indicate evidence for the null hypothesis (and alternative), allowing for confirmation of the hypotheses, which is very informative when trying to confirm the null hypothesis as in the current research, i.e., when testing interaction effects. Incorporating both approaches into a study allows researchers to explicitly address these assumptions, which increases the rigor of the study and allows readers to assess the strength and reliability of the conclusions (Kelter, 2020). So, the Bayesian statistical mixed ANOVAs were implemented as a supplement to null hypothesis significance tests, using JASP 0.17.1 (JASP Team, 2023). The models were compared to the null model (which included the subject variable and random slopes) in each case and calculated the BF_excl values across all models, using its default prior. BF_excls reflect how much more likely it is that the effect does not exist (H0) compared to that it does (H1), given the data. BF_excl values above 1 indicate evidence in favour of H0 over the alternative hypothesis; BFexcl values below 1 indicate evidence in favour of the alternative hypothesis; while values close to 1 indicate that the data are equally likely for null and the alternative hypothesis. Higher values indicate stronger evidence; values of 1–3 indicate weak evidence, 3–10 indicate moderate evidence, and above 10 indicate strong evidence (Jeffreys, 1961).

The purpose of this study was to investigate the reactivity of the triple task method in adults with dyslexia and to compare them with a control group of adults with typical reading skills. It was expected that adults with dyslexia would have difficulty performing the triple task, which could be reflected in writing processes or text quality characteristics. Most previous studies have not confirmed TT reactivity to writing processes or written product in population with typical development (e.g., Penningroth and Rosenberg, 1995; Ransdell, 1995; Piolat et al., 1996) that is, only Wengelin et al. (2014) found more keystrokes and more revisions in the condition without TT. On the other hand, TT reactivity has not been tested in individuals with language disorders such as dyslexia. Moreover, there are no data on TT reactivity measured with KSL measures.

The results of the present study showed no significant interaction effect between group membership and writing condition, suggesting that writing processes and text quality in the group of adults with dyslexia and in the control group are not differentially affected by TT. Furthermore, the fact that no single statistically significant effect was found at p < 0.05 across many comparisons strengthens the evidence that there is no differential reactivity (because multiple comparisons increase the probability that p < 0.05 occurs by chance in one of these comparisons). Thus, with regard to the first objective of this study, it can be assumed that the use of the triple task in adults with dyslexia compared to adults with typical reading skills probably has no effect on the writing processes or product variables measured in this study (as shown in Tables 3 4). It appears that the triple task is a method that is stable enough not to be affected by difficulties that lie in the background of dyslexia and the training of the TT was sufficient to learn the steps of the TT, so that the whole task did not represent a greater additional burden for adults with dyslexia than for the comparison group. However, the Bayesian analyses showed that the evidence for the null hypothesis is rather weak for some interactions, so further research is needed to empower those findings. Also, further research may reveal how reactive the TT is for individuals with dyslexia who write more complex texts and discourse. It is possible that the TT is more responsive to writing processes for more complex texts. For example, Janssen et al. (1996) found that thinking aloud protocols influenced more complex writing processes than simple writing processes (planning level and pause length).

Regarding the second objective, TT reactivity was tested independently of group membership on writing processes and text quality, i.e., TT reactivity was tested independently of dyslexia. The results showed some effects of reactivity. In the condition without TT, the total number of characters per minute was higher and more keys were typed per minute, while the ratio between the total number of words in the final text and the total number of words produced during the writing process showed that more words were deleted in the condition with TT. Wengelin et al. (2014) found that more keystrokes and more revisions were made without TT. Thus, along with our study, they also showed that the triple task may be reactive to some writing processes.

For the third objective, a group of adults with dyslexia was compared to adults with typical reading skills in terms of text quality and writing process. The current study confirmed that adults with dyslexia produced fewer words and sentences in their texts; made significantly more mechanical and grammatical errors when composing texts; produced texts of lower quality in terms of content, coherence, syntax, and vocabulary. Process analysis revealed that adults with dyslexia spent less time actively writing, took more breaks compared to writing time, produced fewer characters per minute, indicating lower transcribing fluency; used fewer keystrokes, had fewer words in the final text compared to the text produced, and typed less. Similar results were found in some previous studies showing that adults with dyslexia produce texts of generally poorer quality and at lexical, structural, and thematic levels compared to controls (Sterling et al., 1998; Connelly et al., 2006; Wengelin, 2007; Torrance et al., 2016). They produce more morphosyntactic, spelling, memory-related, punctuation, and capitalization errors (Sterling et al., 1998; Connelly et al., 2006; Tops et al., 2013; Torrance et al., 2016). Their essays are shorter (Sterling et al., 1998) and their handwriting is less fluent (Connelly et al., 2006).

Finally, some limitations of the study should be discussed. The first limitation stems from the small number of participants, resulting in a lower power of the study. A similar limitation is characteristic of most of the studies presented above. Although dyslexia is not that uncommon, with a prevalence of 5 to 12% in Europe (Dyslexia Compass, 2022), recruiting adults with dyslexia for research is complicated by the fact that this group is no longer in a regular school or health care system and therefore not as easily accessible. In addition, people with dyslexia may even avoid participating in research activities such as reading and writing because they do not want to be exposed to activities that will overwhelm them and make them feel uncomfortable. The second limitation refers to the procedure in which the conditions (TT and no TT) were randomized and not counterbalanced, i.e., the order of the conditions was randomly (not systematically) assigned to the participants. Randomization makes all possibilities equal, thus marginalizing potential serial effects, whereas systematic counterbalancing provides even better control for this type of bias. In addition, picture material was not randomized between conditions, i.e., in one condition, writing text was always elicited by the same prompt. However, the drawings were visually equalized in terms of number of characters and actions, context, and digital painting technique, making it highly unlikely that text quality was related to the properties of the prompt material. Furthermore, the results showed no differences in text quality between the two conditions in either group. Nevertheless, this limitation should be mentioned because randomization of the prompts could eliminate this potential bias. The third potential limitation or problem that should be considered in further research is the fact that participants had to write a narrative essay prompted by pictures. This is a task that could be considered mildly to moderately cognitively demanding. Perhaps a more cognitively demanding discourse, such as an expository text on a particular topic, would reveal more TT reactivity or differences between the groups studied.

When applying a new and complex writing method to a population that has difficulty reading and writing, researchers must be aware of the potential reactivity of the methods used. For example, it is still an open question whether individuals with dyslexia or other language disorders such as Aphasia (MA80.0, ICD-11, WHO, 2019/2021), Developmental language disorder (6A01.2, ICD-11, WHO, 2019/2021) are more sensitive to the TT method than individuals with typical reading or language skills, increasing the likelihood that the TT will be reactive. The results of the current study suggest that TT is unlikely to have differential effects on the writing process or written product in adults with dyslexia compared to adults with typical reading skills. However, further research is needed to strengthen these findings. Nevertheless, the current study and a similar study by Wengelin et al. (2014) provide evidence that the triple task may be reactive in some writing processes regardless of dyslexia. The current study showed that in the no TT condition, the total number of characters per minute was higher, more keys were typed per minute and more words were deleted. When comparing adults with dyslexia and adults with typical reading skills, it was found that adults with dyslexia produced fewer words and sentences in their texts, made more mechanical and grammatical errors, produced texts of lower overall quality, produced fewer characters per minute, used fewer keystrokes, typed less, and had fewer words in the final text than were produced. In addition, researchers need to pay careful attention to TT reactivity, especially in the area of reading and writing disorders, but also when the writing task (or discourse) is strenuous or when new measures of writing process or written product are used in research with TT.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the Centre for Postgraduate Studies Language and Cognitive Neuroscience, University of Zagreb (643–02/14–02/12; 380–130/127–17-13). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

MOT contributed significantly to the concept and design of the work, as it is part of her unpublished doctoral dissertation. She conducted the research, but also analyzed, interpreted, and discussed the data. JKK and RA critically reviewed the work and enriched it with their extensive knowledge and experience in writing articles. They were involved in the entire process of conducting the research and writing the paper. As supervisors, they guided and advised the first author through the extensive research process. All authors contributed to the article and approved the submitted version.