The effect of intuitive- versus deliberative-processing and linguistic versus visual information on auditors’ risk assessment

In recent years, there has been growing recognition of the potential value of using intuitive judgement and decision-making (JDM; Salas et al., 2010). Historically, intuition has often been criticized and associated with errors and biases (Burke and Miller, 1999; Kahneman, 2011). Even within auditing literature, in which deliberative information-processing typically takes precedence, Griffith et al. (2016) theorized that intuitive information-processing might be beneficial when conducting a risk assessment. This study empirically tests this proposition.

Some tasks are better suited to intuitive information-processing, whereas others require more deliberative processing (Hammond, 1996). Deliberative information-processing is slow, analytic and rule-based, whereas intuitive information-processing is typified as fast, holistic and associative (Evans, 2008). This study examines the proposition that task characteristics determine the effectiveness of intuitive versus deliberative processing by investigating how intuitive and deliberative modes influence the outcomes of two essential subtasks within a risk assessment:

1. assessing the risk of material misstatement (RoMM) at the financial statement level; and

2. identifying and documenting risks and controls thoroughly.

We chose these two subtasks because they have different characteristics. Estimating the RoMM at the financial statement level is a more holistic task, for which we expect intuitive processing to be more beneficial. In contrast, identifying and documenting risks and controls represent a more precise task, for which we expect deliberative processing to be more beneficial. In addition, we analyzed the completion time for the overall risk-assessment task.

In current business settings, auditors face visually presented information (Dilla et al., 2010), and audit firms increasingly use dashboards in their audit approach (Eilifsen et al., 2020; Ferdous et al., 2023; Lowe et al., 2018). In practice, visualization is used in various phases of the audit, including during the risk assessment phase (Association of International Certified Professional Accountants (AICPA), 2015; C. Chang and Luo, 2021; B. Wilson and Dennis, 2024). Interestingly, some methods of presenting information align more with certain types of tasks: linguistic information seems to “fit” better with deliberative information-processing, whereas visual information “fits” better with intuitive information-processing (Tuttle and Kershaw, 1998).

We conducted a 2 × 2 between-subjects experiment with 284 auditors, who performed a risk analysis on a fictitious company. The independent variables were processing mode (intuitive vs deliberative) and presentation format (linguistic vs visual). Information-processing was manipulated by using an instruction to process the information either intuitively or deliberatively. As part of the manipulation, we varied the order of the subtasks to stimulate either deliberative or intuitive information-processing. Presentation format was manipulated by presenting the case information either as text or more visually (e.g. graphs).

We found no support of the predicted main effect of deliberative versus intuitive information-processing on the assessment of RoMM. However, we found a significant interaction between information-processing and presentation format. In the linguistic condition, intuitive processing led to higher RoMM estimates than deliberative processing. No difference was observed in the visual condition. This result contradicted our expectations. As expected, deliberative processing led to identifying and documenting more risks and controls than intuitive processing. We further found as expected that intuitive processing resulted in faster task performance. Neither risk identification/documentation nor task time depended on presentation format.

The findings have implications for both audit practice and research. First, building on Fuller and Kaplan (2004), who showed that some audit tasks are better suited to intuitive-processing and others to deliberative-processing, we found indications that these processes have distinct effects within a single task, such as risk assessment. This refines the proposition of Griffith et al. (2021) and Wolfe et al. (2020), who considered this distinction only at the broader task level. Our results suggest that auditors could benefit from recognizing when deliberative information-processing is needed and when intuitive-processing, alongside deliberative-processing, is appropriate for specific subtasks within a task. Targeted training could help auditors understand the distinction between intuitive and deliberative JDM and determine which approach is most effective for a given subtask. Second, Hamdam et al. (2022) theorized that, in contexts in which data visualization is used, intuitive information-processing enhances auditor JDM. However, this claim has not been empirically tested, and there is limited empirical evidence supporting it within the auditing literature. Auditing studies have mainly examined either information-processing mode (Griffith et al., 2015; Hawkins et al., 2021; Wolfe et al., 2020) or presentation mode (Anderson and Mueller, 2005; Backof et al., 2018; Bierstaker and Brody, 2001), with few exploring their combined effects (Rose et al., 2017). The current study addresses this gap and also responds to calls from outside the auditing literature for further research at the intersection of information-processing and cognitive research on data visualization (Padilla et al., 2018). Third, while the use of information visualization has increased in recent years and is generally believed to enhance the quality of auditors’ JDM (Mauludina et al., 2024), the current study demonstrates that it does not necessarily improve JDM for every subtask. Audit firms should, therefore, carefully evaluate the nature of the subtask before developing or implementing such visual tools. Finally, firms and regulators should recognize that while encouraging auditors to approach JDM more deliberatively may lead to more thorough identification and documentation of client risk factors, it does not necessarily increase the perceived client risk level. Therefore, deliberative processing should be used selectively – not as a one-size-fits-all solution for improving audit judgment.

Our study suggests that, for complex tasks, effectively using intuitive or deliberative information-processing depends on not only the nature of the task but also its alignment with the presentation format. Future research should focus on how these factors interact to determine the most effective ways to support auditors in their JDM.

The risk assessment is an important task because it influences the effectiveness and efficiency of audits (Bedard and Graham, 2002; Eining et al., 1997). For this reason, many scholars conduct research into factors that influence auditors’ risk assessment performance (Adikaram and Higgs, 2024; Fay et al., 2015; T. Miller et al., 2012). Griffith et al. (2016) theorized that intuitive information-processing might be beneficial in conducting a risk assessment. Performing a risk analysis consists of several subtasks. In this study, we focus on the outcome of two subtasks (outcome measures):

1. assessing the RoMM at the financial statement level; and

2. identifying and documenting risks and controls.

In addition, we analyze one process measure: the completion time of the risk-assessment task.

The International Standard on Auditing 315 (IAASB, 2023, par. 28) states that auditors are responsible for assessing the RoMM at the financial statement level. The RoMM at the financial statement level represents a holistic risk perception of the client. Auditors must gain in-depth knowledge about the organization, its environment and its internal controls to assess the RoMM (ISA 315.11, A48-A49).

Furthermore, ISA315.38 requires that auditors identify and document risks and controls as part of the risk assessment. The identified risks influence the nature and extent of activities in the audit to address potential sources of misstatement, as described in ISA 315.38 and ISA 330 (IAASB, 2023).

The time taken to perform a task is an important aspect of JDM, because auditors are limited in the time they can devote to a task.

Dual-process theories of reasoning and decision-making assume individuals use two modes of information-processing that influence JDM: deliberative and intuitive information-processing modes (Epstein, 1994; Evans, 2008; Hammond, 1996; Sloman, 1996). Deliberative information-processing is slow, analytic and rule-based, whereas intuitive information-processing is typified as fast, holistic and associative (Evans, 2008).

Some tasks are better suited to intuitive information-processing, whereas others require more deliberative-processing (Hammond, 1996). The sequential analysis of quantitative information and precise JDM are associated with deliberative information-processing (Betsch, 2008b; Dijksterhuis and Nordgren, 2006; Doherty and Kurz, 1996). In contrast, intuitive information-processing is often linked to more holistic, complex JDM (Dane and Pratt, 2007; Dane et al., 2012; Klein, 2003; Sadler-Smith and Sparrow, 2008).

Research on how intuitive information-processing affects auditor JDM is scarce (Griffith et al., 2015; Hawkins et al., 2021; Wolfe et al., 2020), and the results from these studies are mixed. For example, Wolfe et al. (2020) investigated the effect of information-processing on the assessment of impairment indicators and found that intuitive processing outperformed deliberative processing. In contrast, Griffith et al. (2015) examined, among other things, the effect of information-processing on auditors’ ability to recognize unreasonable estimates and to identify inconsistencies and errors in these estimates. They found that deliberative processing outperformed intuitive processing. We argue that these mixed findings might be explained by the type of task: Wolfe et al. (2020) presented relatively straightforward case information and measured participants’ overall perception, which constitutes a more holistic task. In contrast, Griffith et al. (2015) presented more fragmented case information and required participants to identify and justify specific cues, which constitutes a more detailed task. In our experiment, we investigated a holistic subtask and a precise subtask separately, which will be discussed in Section 2.4.

Information can be presented in a more linguistic or a more visual format. The linguistic format contains natural narratives and grammar. The visual format includes charts, images (Tang et al., 2014), flowcharts (Bierstaker and Brody, 2001), visual metaphors (Eppler and Aeschimann, 2009) and other visual symbols. Studies investigating presentation format have used different labels: “diagrammatic” versus “linguistic information” (Dunn and Gerard, 2001), “verbal” versus “pictorial presentation” (Holbrook and Moore, 1981) and “graphical” versus “tabular information” (Vessey, 1991). In this paper, we use the labels “linguistic information” and “visual information.”

In practice, visualization is used in various phases of the audit, including during the risk assessment phase (Association of International Certified Professional Accountants (AICPA), 2015; C. Chang and Luo, 2021; B. Wilson and Dennis, 2024). For example, Association of International Certified Professional Accountants (AICPA) (2015) described that visualization “[…]can help the auditor understand the business, identify anomalous patterns or outliers, and ultimately plan the audit” (p. 108). Both financial and non-financial information can be visualized in dashboards used by auditors (Salijeni et al., 2021). For example, the visualization of process mining offers insights into process controls like workflow and segregation of duties (Jans et al., 2014).

According to the cognitive fit theory (Vessey, 1991), certain ways of presenting information are better suited to specific types of tasks. Visual information is preferable for more holistic tasks that require synthesizing multiple pieces of information. In contrast, linguistic information is more effective for identifying or recalling more precise information (Anderson and Mueller, 2005; Umanath et al., 1990; Vessey, 1991). A “fit” between task and presentation format supports efficient and effective JDM, while a “misfit” may hinder effective processing and decreased accuracy (Vessey, 1994). Despite being over 30 years old, Vessey’s theory remains widely used in more recent auditing literature (Backof et al., 2018; Baaske et al., 2025; Holt and Loraas, 2021; Sihombing et al., 2023).

The cognitive fit theory proposed by Vessey (1991) has been extended to include information-processing (Padilla et al., 2018; Tuttle and Kershaw, 1998). For example, Tuttle and Kershaw (1998), in a management-judgment context, predicted and found that holistic JDM is enhanced by graphical presentations, whereas deliberative JDM benefits from tabular formats. The underlying premise is that deliberative information-processing aligns or “fits more closely with linguistic information, whereas intuitive-processing is better suited to visual information” (Epstein, 1994; Hogarth, 2001; Inbar et al., 2010; Sauter, 1999; Sloman, 1996). In the auditing domain, Hamdam et al. (2022) proposed that using data visualization enhances auditor JDM through intuitive information-processing. However, this proposition has yet to be empirically tested.

As discussed above, both dual-process theories and cognitive fit theory typically examine effects at the task level. In this study, we investigate how these effects manifest at the subtask level within a risk analysis task. We chose the RoMM at the financial statement level (ISA 315.30) and the identification and documentation of risks and controls (ISA 315.38) as focal tasks. The RoMM at the financial statement level consider risks that relate pervasively to the financial statements as a whole and may potentially affect many assertions (ISA 200.A38) ^[1]. We selected these two subtasks because they differ in nature: the first represents a holistic subtask, whereas the second represents a precise subtask. We, therefore, expect the modes of information-processing to have different effects across the two subtasks.

Assessing RoMM at the financial statement level is challenging because of the large volume of information auditors must process. Intuitive processing may, therefore, offer advantages. Finucane et al. (2000, p. 3) stated the following:

This quote provides several reasons why intuitive information-processing positively contributes to a complex task, such as assessing the RoMM at the financial statement level. First, intuitive information-processing seems better equipped than deliberative-processing to develop a holistic understanding of different, and sometimes even contrasting, information cues (Dane and Pratt, 2007; Dijksterhuis and Nordgren, 2006; Inbar et al., 2010; Klein, 2003; Schooler et al., 1993). Furthermore, intuitive information-processing increases the likelihood of recognizing patterns that may not be consciously identified (Bowers et al., 1990). Second, intuitive information-processing is associated with affect (Epstein et al., 1996); therefore, auditors may use feelings as warning signals (somatic markers; Damasio, 1999) that are not readily available in the conscious mind (Sadler-Smith and Shefy, 2007). Intuitive information-processing has a lower threshold for risk factors than deliberative information-processing; as a result, deliberative processing is expected to require more information before responding (Darlow and Sloman, 2010). Consequently, when auditors process information intuitively, they focus on and attach greater weight to risk factors (red flags). Third, compared with deliberative information-processing, intuitive-processing offers higher processing capacity (Dijksterhuis et al., 2006; Salas et al., 2010). Therefore, auditors who use intuitive information-processing should be able to process and integrate larger amounts of information than auditors who use deliberative processing. In summary, auditors who rely on intuitive information-processing are expected to recognize risky patterns more quickly and assign greater weight to risks through affective cues. Intuitive processing also enables them to assess a larger volume of risk information in a more holistic manner. Based on these arguments, we expect the following:

As described in Section 2.3, the cognitive fit theory proposed by Vessey (1991) has been extended to include information-processing (Padilla et al., 2018; Tuttle and Kershaw, 1998). The assumption is that deliberative information-processing aligns or “fits” more closely with linguistic information, whereas intuitive-processing is better suited to visual information (Epstein, 1994; Hogarth, 2001; Inbar et al., 2010; Sauter, 1999; Sloman, 1996). Hypothesis H1a states that assessing risks is supported by intuitive information-processing. Consistent with the matching principle (Tuttle and Kershaw, 1998; Vessey, 1991), we expect intuitive processing to benefit more from visual information than from linguistic information. This leads to the following hypothesis:

Another relevant subtask in auditors’ risk assessment is the identification and documentation of risks and controls. Note that this subtask is limited to the documentation of the controls and their effectiveness as required in ISA 315.38; we abstract from the subtask of assessing and evaluating the controls themselves (ISA 315.A196). We propose that deliberative information-processing is more beneficial than intuitive-processing for the identification and documentation of the number of risks and controls for two primary reasons. First, as previously discussed, deliberative information-processing is well-suited to sequential, detailed JDM. We argue that identifying and documenting risks and controls is a task better suited for step-by-step (sequential) execution. Consequently, deliberation may lead to the identification and documentation of more risk and control factors. Second, it is more difficult to articulate the rationale underlying judgments when relying on intuitive processing (Betsch, 2008b; Sadler-Smith, 2008; Sinclair and Ashkanasy, 2005). As a result, in the intuitive processing mode, auditors may rely on certain controls in practice but fail to articulate or document them explicitly. Moreover, auditors using an intuitive mode are likely to focus primarily on controls they perceive as most relevant, whereas auditors using a deliberative mode are more likely to evaluate both relevant and less relevant controls. This leads to the following hypothesis:

As described in Section 2.3, deliberative information-processing is supported by a linguistic presentation format (Tuttle and Kershaw, 1998). Therefore, we expect that the identification and reporting of risks and controls will benefit from deliberative information-processing, particularly when the risk information is presented in a linguistic format. Therefore, we expect the following:

Auditors face time constraints when performing tasks. Therefore, the time used to conduct a risk analysis is crucial. Dual-process theories describe deliberate information-processing as effortful and slow (for an overview, see Evans, 2008). Intuitive information-processing, on the other hand, is typified as fast and holistic (Dane and Pratt, 2007). Therefore, we hypothesize the following:

As described in Section 2.3, a fit between task and presentation format enhances efficiency and leads to faster JDM (Tuttle and Kershaw, 1998; Vessey, 1991). Consequently, intuitive information-processing is expected to be faster than deliberative-processing, particularly when auditors receive information visually. This point leads to the following hypothesis:

We investigated the effects of information-processing mode and presentation format. The study included two information-processing modes (deliberative and intuitive) and two presentation formats (linguistic and visual), resulting in a 2 × 2 between-subjects design, with participants randomly assigned to one of the four groups. The experiment was implemented using Qualtrics (Qualtrics, Provo, UT).

The first independent variable, information-processing, was manipulated through direct instructions, adapted from procedures used in previous studies (Jordan et al., 2007; T. Wilson and Schooler, 1991). In the intuitive condition, participants were instructed to rely on their intuition, making judgments spontaneously based on first impressions and feelings. In contrast, participants in the deliberative condition were instructed to think carefully and disregard their initial impressions (in line with Dane et al., 2012). As part of the manipulation, we also varied the order of the subtasks. Inspired by previous research (Jordan et al., 2007; T. Wilson and Schooler, 1991), participants in the deliberative condition were required to document the risks and controls before assessing the RoMM. This approach aligns with findings from Hawkins et al. (2021), suggesting that requiring auditors to justify their judgments shifts information-processing toward a more deliberative approach. In the intuitive condition, participants documented the risks and controls after assessing the RoMM.

The second independent variable, presentation format, was manipulated by varying how information was displayed. As discussed in Section 2.3, linguistic formats contain natural narratives and grammar, whereas visual formats include charts, images, visual metaphors and other visual symbols. In the current study, the linguistic condition consisted primarily of plain text and tables. In the visual condition, text and tables were replaced with graphs, infographics and visual icons wherever possible. An example of the linguistic and visual condition is provided in the Supplementary Material.

The participants received a case about a hypothetical trade organization, ENF, based on real company data ^[2]. The ENF case contained diagnostic (risk and control factors) and non-diagnostic (irrelevant factors) information. The diagnostic information included details about the general economic situation, management turnover rate, management reputation, bonus system, IT auditor findings and financial ratios. Based on earlier research (S. Chang et al., 2008), we included 24 risk and control factors in the case scenario: 11 risk factors were classified as risks, 11 as controls and 2 were deemed ambiguous (capable of being interpreted as either risks or controls). The objective was to construct a case reflecting an average risk profile. The cases included various risks at the financial statement level, such as the appointment of a new Financial Director after the predecessor’s retirement, the introduction of a bonus system focused on financial growth and IT audit findings on deficiencies in authorization management within the financial system. The controls included entity level controls like general IT controls and internal audit department. The participants were instructed to imagine themselves as an in-charge senior audit manager responsible for conducting a risk assessment for a case.

At the start of the experiment, the participants received condition-specific instructions on processing the case. They then reviewed the case information which was divided across eight screens on the computer, without the possibility to go back to previous pages. Afterward, the participants estimated the RoMM and documented the risks and controls, with question order varying by condition.

Following the risk assessment, participants completed a questionnaire covering their background (professional skepticism, preference for intuition and deliberation, experience and gender), participants’ perceptions of the case and their understanding of the manipulation.

The participants were Dutch auditors recruited using the snowball-sampling technique. The initial sample consisted of 48 students from the Master of Science in Accountancy program at a Dutch University. All the students were affiliated with audit firms and studied part-time (one day a week). The students were asked to participate voluntarily in this experiment: 47 students agreed to participate and provided informed consent [2^[[3]]]. As part of this project, the students were asked to recruit at least seven potential participants from their firm, at the junior manager level or higher. Students were instructed on what experiment information could be shared and signed a declaration of compliance. There was no penalty nor any disadvantage for students who failed to find seven participants. Potential participants received an invitation email from the first author, with a short introduction and a personal link to enter the study. Potential participants who did not respond received a reminder via email.

This study was conducted using Qualtrics (Qualtrics, Provo, UT). Leiby et al. (2021) raised concerns about using online platforms in accounting research. To address these, we invited only individuals with audit firm email addresses, sent unique single-use study links and required participants to confirm they had received the invitation directly from the first author.

The students recruited a total of 446 participants who agreed to take part in the experiment. A total of 436 external auditors completed the experiment. In the introduction of the experiment, participants were instructed to complete the experiment in one session. Those participants who indicated they did not complete the experiment in one session (n = 120) were excluded from the analyses ^[3]. Additionally, another seven participants were excluded: six because of failure to follow procedures and one because of experiencing technical difficulties. We excluded 25 participants who reviewed the case in under 200 s, as this was below our estimated minimum time for a basic review. After these exclusions, 284 participants remained available for the analysis.

Dependent variables: The assessment of the Risk of Material Misstatement (RoMM_ASMT) at the financial statement level for the case ENF was measured using the following instruction: Indicate your provisional risk assessment on the scale below given the aforementioned information on ENF. The participants could indicate their assessment using a horizontal slider (1–100) with endpoints labeled Low and High and a midpoint labeled Medium (Wright and Bedard, 2000). The numerical scale associated with the slider positions was undisclosed to the participants.

To identify and document the risk and control factors, we asked the participants to report the Risk Factors and the Control Factors they identified in the ENF case. This request was achieved through two open-ended questions: What information about ENF has a risk-increasing (decreasing) effect in your opinion? We analyzed the written responses to these questions and categorized them into risk factors and control factors ^[4]. Finally, we counted the number of risk and control factors that match the factors we included in the ENF case. Those two variables were used in the analyses.

The Time Used by a participant to conduct the risk assessment includes the duration spent reviewing the ENF case material, assessing the RoMM and documenting the risks and controls, as recorded by Qualtrics.

Covariates: Prior research has indicated the following variables are related to auditor JDM: gender, PID, experience and skepticism (Hummel et al., 2025). We measured skepticism using the 30-item Hurtt Professional Skepticism Scale (HPSS; Hurtt, 2010; Quadackers et al., 2014). An example item is the following: I often take statements from other people without thinking. In the current study, the 30 items explained 19.9% of the variance in skepticism, and Cronbach’s alpha of the scale was acceptable (α = 0.79). Higher scores on the scale indicate higher levels of skepticism. We measured preference for intuition and deliberation using Betsch’s PID scale (Betsch, 2004, 2008a). The scale comprised nine items measuring preference for intuition (PID-I), such as, I pay close attention to my deepest feelings, and nine items measuring preference for deliberation (PID-D), such as: Before making decisions, I think them through. Each item was scored on a five-point scale, ranging from Very much disagree (1) to Very much agree (5). In the current study, the nine PID-I items explained 37.1% of the variance in the PID-I, and Cronbach’s alpha of the scale was acceptable (α = 0.78). The nine PID-D items explained 35.8% of the variance in the PID-D; Cronbach’s alpha of the scale was acceptable (α = 0.75). We measured Experience by asking participants for their number of years of experience in auditing. Gender was assessed by asking participants to indicate their gender, with the response options: (0) male and (1) female.

We conducted separate ANOVAs on participant characteristics (preference for intuition, preference for deliberation, skepticism, experience and gender) to test randomization across the four groups. As shown in Table 1, Panel B, no significant differences emerged, indicating successful randomization.

Case characteristics: We measured the participants’ perceptions of the ENF case by following Fuller and Kaplan’s (2004) approach. Single-item, ten-point scale measures captured perceived Complexity, Realism and Ability to perform the task. For example, complexity was measured with: How do you assess the complexity of the case? Simple (1)–Complex (10). Panel C of Table 1 presents the results. The participants perceived the task as moderately complex (M = 5.03; SD = 1.28) and realistic (M = 7.89; SD = 1.30). Moreover, they believed they could perform this task well, given the average score regarding Ability (M = 7.57; SD = 1.78).

An overview of all dependent variables, covariates and case characteristics is provided in the  Appendix.

Manipulation check: We implemented two manipulation checks to assess the extent to which the participants performed their JDM in line with the manipulations. For the information-processing manipulation (mcIP), we used a four-item instrument inspired by Zimbelman (2014) and Wolfe et al. (2020). Each item was scored on a ten-point scale, with response scales specific to each item. An example item is the following: I have made the risk assessment based on: my feelings (1)–intellect (10). A higher composite score on the four items indicated more deliberative processing; the items accounted for 54.1% of its variance. To test whether our experimental manipulation worked, we performed an independent-samples t-test. The non-tabulated manipulation check mcIP indicated a significant difference, t(282) = 3.45, p < 0.001, in the expected direction, between the intuitive condition (M_mcIP = 5.04; SD_mcIP= 1.32) and deliberative condition (M_mcIP= 5.61; SD_mcIP= 1.47). This result indicates our manipulation was successful.

To assess the presentation format manipulation (mcPF), we asked how the participant perceived the presentation format, with the following response options: (1) mainly text, (2) a combination of visual information and text or (3) mainly visual information. A Kruskal–Wallis test showed a significant difference in manipulation-check scores between conditions, H(1) = 120.61, p < 0.001. Participants in the visual condition had a higher mean rank (MR_mcPF = 188.11) than those in the text condition (MR_mcPF = 96.24), indicating that the manipulation was successful.

Table 1 contains the descriptive statistics of the variables in the analyses. Panel A displays the dependent variables. On average, the participants estimated the ROMM_ASMT to be 62.51 (SD_{RoMM_ASMT}= 14.54) on a scale ranging from 1 to 100. This suggests that the participants perceived the case as having a slightly above-average risk profile. On average, the participants identified and documented nearly seven risk factors and control factors. This finding aligns with the general idea that working memory capacity is limited to approximately seven items (Cowan, 2015; G. Miller, 1956; Norris and Kalm, 2021). The participants reported more Risk Factors (M_RF = 4.39; SD_RF = 1.99) than Control Factors (M_CF = 2.42; SD_CF = 1.39). On average, the participants took 14.11 min (M_TU = 846.38 s; SD_TU = 545.09 s) to read the case material and to answer the risk assessment questions. Panel B of Table 1 displays the individual characteristics. On average, the participants had 9.46 years (SD = 7.49 years) of auditing experience.

To test Hypotheses H1a and H1b, we conducted a 2 × 2 ANCOVA analysis with the RoMM_ASMT as the dependent variable and Experience, Gender, PID-D, PID-I and HPSS as covariates ^[5]. The results, presented in Table 2, include descriptive statistics in Panel A and the accompanying ANCOVA table in Panel B. No significant main effect of Information-Processing (H1a) was found, F(1, 275) = 0.21 and p = 0.650. The interaction between Information-Processing and Presentation Format (H1b) was significant, F(1, 275) = 4.20 and p = 0.041. The interaction plot in Figure 1 illustrates the direction of this interaction. Judging from the interaction plot, the difference between RoMM_ASMT_intuitive and RoMM_ASMT_deliberative was larger in the linguistic condition, in disagreement with Hypothesis H1b.

To assess in which presentation condition the difference between processing conditions was significant, we performed Bonferroni-corrected pairwise comparisons (simple effects, non-tabulated). These analyses indicated that, in the visual condition, there was no significant difference, F(1,280) = 1.68 and p = 0.20, between the intuitive condition (M_{RoMM_ASMT} = 61.11 SD_{RoMM_ASMT} = 13.07) and the deliberative condition (M_{RoMM_ASMT} = 64.27; SD_{RoMM_ASMT} = 14.47). In the linguistic condition, we found a significant difference, F(1,280) = 4.58 and p = 0.03, between the intuitive condition (M_{RoMM_ASMT} = 64.85; SD_{RoMM_ASMT} = 14.32) and the deliberative condition (M_{RoMM_ASMT} = 59.60; SD_{RoMM_ASMT} = 16.19), in the expected direction.

Hypothesis H1a – intuitive information-processing results in a higher risk estimate than deliberative-processing – was not supported. Furthermore, Hypothesis H1b – intuitive information-processing results in a higher risk estimate than deliberative information-processing, especially in the visual condition – was not supported. However, a significant interaction effect emerged. Simple effects analyses revealed that, contrary to our expectations, the difference between RoMM_ASMT_intuitive and RoMM_ASMT_deliberative appeared only in the linguistic condition. In this condition, intuitive participants estimated RoMM significantly higher than deliberative ones, partially supporting H1a.

To test Hypotheses H2a and H2b, we conducted a 2 × 2 ANCOVA analysis with the number of client Risk Factors and Control Factors as the dependent variables and Experience, Gender, PID-D, PID-I and HPSS as covariates. We studied the risk and control factors separately.

The results of the analysis with Control Factors as the dependent variable are presented in Table 3 the descriptive statistics in Panel A and the ANCOVA table in Panel B. The results indicate a significant main effect of Information-Processing (H2b), F(1, 275) = 31.58 and p < 0.000. The participants reported more Control Factors (CF) in the deliberative condition (M_CF= 2.91; SD_CF = 1.48) than in the intuitive condition (M_CF= 2.01; SD_CF = 1.17). The interaction between Information-Processing and Presentation Format (H2b) was not significant, F(1, 275) = 0.88 and p = 0.349.

Table 4 lists the results of the analysis with Risk Factors as the dependent variable. The results reveal a significant main effect of Information-Processing (H3a), F(1, 275) = 24.79 and p < 0.001. The participants reported more Risk Factors (RF) in the deliberative condition (M_RF = 4.98; SD_RF = 2.05) than in the intuitive condition (M_RF = 3.92; SD_RF = 1.82). The interaction between Information-Processing and Presentation Format (H3b) was not significant, F(1, 275) = 3.57 and p = 0.060.

Hypothesis H2a – deliberative information-processing results in the reporting of more risks and controls than intuitive information-processing – is supported for both the risks and controls. Hypothesis H2b – deliberative information-processing results in the reporting of more risks and controls than intuitive information-processing, especially in the linguistic condition – is not supported for either the risks or the controls.

To test Hypotheses H3a and H3b, we conducted a 2 × 2 ANCOVA analysis with Time Used (TU) to complete the risk assessment as the dependent variable and Experience, Gender, PID-D, PID-I and HPSS as covariates ^[6]. Table 5 present the results: Panel A the descriptive statistics, and Panel B the accompanying ANCOVA table. There was a significant main effect of Information-Processing, F(1, 274) = 14.20 and p < 0.000. The participants in the intuitive condition (M_TU = 738.73; SD_TU= 404.86) were, on average, 241.8 s (approximately 4 min) faster than the participants in the deliberative condition (M_TU = 980.52; SD_TU=658.06). The interaction between Information-Processing and Presentation Format was not significant, F(1, 274) = 0.09 and p = 0.766.

The Time Used is the time each participant spent on the risk assessment. This includes the time used to review the ENF case (Time Factors) and the time to estimate the RoMM and to report the risks and controls (Time Answers). In post hoc analyses, we examined whether the effect of information-processing on Time Used was also evident for Time Factors and Time Answers separately. We conducted an analysis similar to that for Time Used, substituting Time Used with Time Factors and Time Answers, respectively (non-tabulated).

The analysis with Time Factors as the dependent variable resulted in a non-significant model (p = 0.113). This indicates that there was no difference between the conditions in the speed with which participants read the information. In the analysis with Time Answers (TA) as the dependent variable, we found a significant main effect of Information-Processing, F(1, 274) = 35.14, p < 0.001 and η² = 0.114. The participants in the intuitive condition (M_TA = 196.64; SD_TA = 137.03) were, on average, 3 min and 33 s faster answering the questions than the participants in the deliberative condition (M_TA = 409.39; SD_TA = 419.63). Therefore, the participants in the deliberative condition were not slower in reviewing the ENF case, but they did spend more time assessing the RoMM and documenting risks and controls.

Hypothesis H3a – intuitive information-processing results in a faster risk assessment than deliberative information-processing – is supported for Time Use. The post hoc analyses indicated this difference in task performance can be attributed to faster JDM. Hypothesis H3b – intuitive information-processing results in a faster risk assessment than deliberative information-processing, especially in the visual condition – is not supported.

Possibly, the extent of experience with the risk assessment task is affecting the results. This is suggested by Anderson and Mueller (2005), who found an interaction between presentation format and experience. Therefore, we reran the main analyses excluding the part-time students, resulting in a sample of n = 250 (results untabulated). Furthermore, we analyzed the effect of experience in sub-groups of low experience (<6 years, n = 137), medium experience (6–12 years, n = 71) and high experience (>12 years, n = 76). Only in the medium experience group we found significant results supporting those reported in Table 1. For both analyses, we found no support for the idea that experience is driving the results.

Table 6 presents a summary of the hypotheses, statistical results, hypothesis testing outcomes, interpretations and post hoc explanations. Post hoc explanations for unsupported findings are discussed in more detail in the discussion section (Section 5 .1).

We investigated the effect of information-processing mode (intuitive vs deliberative) and presentation format (linguistic vs visual) on auditors’ JDM in a risk-assessment task. We focused on the outcome of two subtasks (outcome measures):

1. assessing the RoMM at the financial statement level; and

2. identifying and documenting risks and controls.

Additionally, we analyzed one process measure: the completion time of the risk-assessment task.

We hypothesized that auditors encouraged to process information intuitively would estimate a higher RoMM than those encouraged to process it deliberatively. Contrary to our expectations, we found no differences in the assessment of the RoMM between the deliberative and intuitive information-processing modes. We hypothesized that auditors encouraged to process information intuitively would estimate the RoMM higher than those encouraged to process the information deliberatively, especially when exposed to visual information. We found a significant interaction between information-processing and presentation format. In the linguistic condition, intuitive processing led to higher RoMM estimates than deliberative processing. No such difference was observed in the visual format. This result contradicts our expectations but partially aligns with our first hypothesis. We consider the following ad hoc explanation for the absence of the anticipated interaction effect. Prior research indicates that intuitive processing often depends on visual familiarity (Jones, 2015; Sanjari et al., 2017). Consequently, limited familiarity with the visuals used in this study may have weakened the potential advantages of visualization. The visual presentation format applied in the experiment was uniquely designed for this study and may not align with those auditors typically encounter in professional practice. Vera‐Muñoz et al. (2001) argued that auditors’ underlying knowledge structures may not be fully activated when they engage with unfamiliar visualizations, further limiting their ability to benefit from them. The use of visuals may have disrupted intuitive information-processing in the visual condition. Linguistic information alone requires more processing capacity than linguistic information supported by visuals (Mayer, 2005). This higher processing demand may have led auditors to benefit more from their intuition in the linguistic condition than in the (unfamiliar) visual condition. Thus, a certain degree of visualization unfamiliarity and unequal processing capacity requirements in all conditions are aspects that might explain the unexpected results and can be considered in future research.

Furthermore, we hypothesized and found that auditors encouraged to process the information deliberatively identified and reported more risks and controls than those encouraged to process the information intuitively. This observation emphasizes the direct relationship between careful consideration and the thorough identification and documentation of audit evidence by auditors. Contrary to expectations, no significant interaction emerged between information-processing and presentation format in identifying and documenting risk factors. A post hoc explanation can possibly be found in the study of Speier (2006). For tasks requiring the identification or recall of precise information, Speier found similar decision accuracy and completion times across visual and linguistic formats. Thus, the presentation format used in our experiment had less impact than expected on tasks that require identifying and recalling precise information, such as identifying and documenting risks and controls.

We hypothesized and found that auditors encouraged to process the information intuitively completed the risk-assessment task faster than those encouraged to process the information deliberatively. As an additional analysis, we separated the information review and JDM phases. The results indicate the participants in the intuitive condition reviewed the information just as quickly as those in the deliberative condition. In the JDM phase, however, auditors in the intuitive condition were faster but identified and documented fewer risk factors. We did not find an interaction between processing mode and presentation format for time spent on the task. One possible reason is that, as discussed above, auditors were unfamiliar with the visual format, which may have slowed intuitive processing.

The current study offers a preliminary insight into how intuitive and deliberative information-processing influence auditors JDM, considering the role of information presentation. For (sub)tasks that require more sequential and detailed decision-making, such as the identification and documentation of risk factors, deliberative information-processing consistently outperformed intuitive-processing. This effect was observed regardless of the information presentation format. In contrast, for more holistic (sub)tasks that involve integrating multiple pieces of information – such as assessing the RoMM at the financial statement level – the relationship was more complex. Intuitive information-processing resulted in a higher RoMM estimate, but only in the linguistic condition. This finding suggests that, in cases with an average risk profile, intuitive-processing may lead to more skeptical judgments. However, its effectiveness appears to depend on not only the nature of the (sub)task but also the presentation format. Our findings show that intuitive and deliberative processing affect risk assessment subtasks differently, and in some subtasks, their effect also depends on the presentation format. Therefore, deliberative processing should be used selectively – not as a one-size-fits-all solution for improving audit judgment.

Our experiment has limitations that may guide future research to strengthen our conclusions. First, the visuals used in this study may have differed from those typically used by audit firms. This difference may have limited their effectiveness in supporting intuitive information-processing. Future research could address this issue by allowing auditors to familiarize themselves with a specific presentation format before the experiment or by explicitly examining familiarity. For example, familiarity could be incorporated as a manipulated variable in the study design. Second, although a direct instruction to manipulate information-processing is an established technique, this technique cannot fully isolate intuitive versus deliberative information-processing. Third, the experiment used visualizations, such as charts and images. In practice, however, other forms of visualization may be used, including scatterplots, trend lines or bubble charts (C. Chang and Luo, 2021), as well as flowcharts (Bierstaker and Brody, 2001). These alternatives fall outside the scope of this study and may lead to different findings. Fourth, to ensure informational equivalence between the linguistic and visual formats, the visual condition also included textual elements. This design choice may have diluted the contrast between formats and potentially reduced the strength of any interaction effects. Future research could strengthen the distinction between textual and visual formats to better isolate their respective effects. Finally, the case used in this experiment had an average risk profile, which restricts generalizability across client risk profiles. Future research could explore this aspect by designing an experiment that manipulates the risk profile in addition to information-processing.

The authors appreciate the assistance from the Nyenrode students who helped collect the data and the auditors who participated in this study. The authors thank Martijn van de Berg, Diane Breesch (discussant), Nicolien Fiere, Anna Gold, Christopher Koch, Justin Leiby (discussant) and Annemarie Wennekers for their valuable feedback. The authors have received valuable comments from participants of the PhD workshop of the European Auditing Research Network conference in Parma (2019), participants of the Accounting Research Day in Brussels (2022) and participants of the Academic Brown Bag Session, Nyenrode Business University in Breukelen (2022), workshop participants of from the Internal Audit Department of the Ministry of Finance in The Hague (2023). The authors are grateful to the anonymous reviewers and Jie Zhou (Editor-in-Chief) for their helpful comments and suggestions.

The supplementary material for this article can be found online.