The use of neurocognitive methods in assessing health communication messages: A systematic review

Abstract

We review 20 studies that examined persuasive processing and outcomes of health messages using neurocognitive measures. The results suggest that cognitive processes and neural activity in regions thought to reflect self-related processing may be more prominent in the persuasive process of self-relevant messages. Furthermore, activity in the medial prefrontal cortex, the superior temporal gyrus and the middle frontal gyrus were identified as predictors of message effectiveness, with the medial prefrontal cortex accounting for additional variance in behaviour change beyond that accounted for by self-report measures. Incorporating neurocognitive measures may provide a more comprehensive understanding of the processing and outcomes of health messages.

Keywords

behaviour change event-related potentials functional magnetic resonance imaging health communication message processing

Health campaigns focus upon raising awareness of the negative consequences associated with risk-taking behaviours, preventing future health problems by promoting healthy behaviours and/or detecting potential life-threatening illnesses. Given the high costs associated with designing health campaigns and the potentially important role that these advertisements may have on improving population health, it is imperative that health advertisements are perceived by the intended audience to be persuasive and, in turn, lead to behaviour change.

Research which has examined the persuasive effects of health messages has typically relied upon self-report measures of message acceptance, such as attitudes and behavioural intentions (Ajzen, 1991) and ratings of perceived message effectiveness (Dillard et al., 2007). Although theoretical (e.g. theory of planned behaviour (TPB); Ajzen, 1991) and empirical evidence supports these constructs as influencing subsequent behaviour (e.g. Fortier et al., 2009), there remains a substantial variance in behaviour which is unaccounted for by such constructs. In order to increase our understanding of the underlying processes involved in message processing and subsequent acceptance (or rejection) of health communication messages, neurocognitive measures, such as event-related potentials (ERPs) and functional magnetic resonance imaging (fMRI), could be used to complement the existing self-report measures. Neurocognitive measures overcome many of the limitations of self-report (e.g. social desirability) and have been found to explain additional variance in behaviour change, above and beyond self-report measures of attitudes and intentions (e.g. Falk et al., 2011). Furthermore, neurocognitive measures may offer an alternative neural network–based explanation for the processes involved in message persuasion and consequently help to advance theoretical knowledge in this area.

Only recently have researchers started to examine neurocognitive activity in response to health messages in order to assess the role of early attentional biases and more complex processing differences (e.g. Falk et al., 2011; Kessels et al., 2014). Two neurocognitive measures, ERPs and fMRI, represent the focus of this review.¹ ERPs are a high-temporal resolution method that uses the averages of event-linked (e.g. presentation of an auditory or visual stimulus) changes in neural electrical potentials close to the scalp, in order to assess attention and/or cognitive processing of those events (Duncan et al., 2009). ERP components can differ in terms of polarity (positive vs. negative), latency (peak of the wave) and scalp distribution, with each component reported to be linked to separate underlying neural processes. Two examples of ERP components are the N100 and the P300. The N100 component (a negative potential occurring approximately 100 ms after stimulus onset) is thought to reflect earlier attentional processes (Coull, 1998). The N100 has been reported to be more pronounced over the frontal and central cortical regions and is observed when stimuli are attended to compared to unattended stimuli (Coull, 1998). The P300 component (positive potential approximately 300 ms following stimulus presentation) is more pronounced over the central and parietal neural regions and has been reported to be associated with later higher order cognitive processes such as controlled attentional processes and working memory (e.g. matching a stimulus on screen with that presented three stimuli previously; Polich, 2007).² fMRI is high in spatial resolution and can be used to assess the changes in the blood oxygen level-dependent (BOLD) signal, which allows researchers to identify specific brain structures involved in processing (e.g. Cui et al., 2011). In a health communication context, fMRI can be used to assess the subcortical structure involvement associated with message processing, which is important given that previous research has reported that persuasion involves both social and emotional processes, requiring activation of both cortical and subcortical areas (e.g. Falk et al., 2011, 2015).

These neurocognitive measures may be particularly useful in advancing understanding of health messages as they can measure message processing in real time via moment-by-moment, continuous recording. For instance, considering that cognitive and emotional responses are likely to vary throughout advertisement exposure, depending on the moment-by-moment context of an advertisement, neurocognitive measures enable the measurement of those changes that observers (and even individuals themselves) may not be consciously aware of, at the exact time that they are occurring. Thus, neurocognitive measures may offer additional insight not otherwise available via alternative measurement techniques.

With a now solid and growing evidence base in the health communication field, it is timely to review the body of literature to evaluate the feasibility of neurocognitive measures for the assessment of processing and persuasive outcomes of health communication. Therefore, the purpose of this article was to systematically review the scientific literature which has applied neurocognitive techniques to examine message processing and their contribution to message persuasion/behaviour change with a focus in the health advertising context.

Method

Search strategy

This review comprised articles published in PsycINFO, PubMed, Web of Science and ProQuest Psychology Journals until September 2015. The following search strategy was used in all databases: ((neural or neuro* OR functional magnetic resonance imaging OR fMRI OR Event related potential OR ERP OR electroencephalography OR EEG OR functional near infrared spectroscopy OR fNIRS OR magnetic resonance imaging OR MRI OR Magnetoencephalography OR MEG) AND (health message OR public service announcement OR PSA OR communication message) AND (persuasion OR attention OR processing OR acceptance OR effectiveness OR intention OR behavio*r change)). In the ProQuest Psychology Journals database, the following limits were also applied: scholarly peer-reviewed journals, English, humans and sources from 1997 to 2015. No limits were applied to the remaining databases at this initial search stage.

Inclusion and exclusion criteria

The Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Liberati et al., 2009) were used for the study selection process (see Figure 1 and online supplement for further information). One author was contacted to clarify the types of activities that were included in text-based messages (i.e. Falk et al., 2010). It was verified by this author that not all arguments included in the messages were related to health behaviours. However, this article was retained as some of the arguments included in the messages related to health and pro-social type issues (e.g. dental hygiene/flossing, blood donation). This review consisted of a final set of 17 articles (Chua et al., 2009, 2011; Cooper et al., 2015; Dinh-Williams et al., 2014; Falk et al., 2009, 2010, 2011, 2015; Kessels et al., 2010, 2011, 2014; Langleben et al., 2009; Ramsay et al., 2013; Ruiter et al., 2006; Seelig et al., 2014; Wang et al., 2013; Weber et al., 2015).

Figure 1.

Flowchart of the inclusion procedure.

Data extraction

An excel spreadsheet was used to extract the relevant data (see online supplement for a list of the items that were extracted). The co-authors reviewed the extracted data and any differences in interpretations were discussed and mutually agreed upon. Of the 17 articles included in this review, Kessels et al.’s (2014) article reported on two studies and Falk et al.’s (2009) article included three studies (each of these studies comprised a separate sample). Therefore, a total of 20 studies (5 ERP and 15 fMRI studies) were reviewed. The extracted data were grouped into three categories: study characteristics, ERP studies and fMRI studies.³

Results

Study characteristics

All studies were published in English between 2006 and 2015. The studies comprised a total of 742 participants (59% female), with 73 of these participants excluded from final data analyses. A large proportion of those 73 participants (67%)⁴ were excluded due to excessive artefacts (noise) in neurological data, technical problems, mistakes and/or poor performance on the experimental tasks (see Table 1 for study details).

Table 1.

Study and sample characteristics of the included articles.

Study	Study characteristics				Sample characteristics
	Measurement technique/area of interest	Design	Message medium	N	M age (SD)Age range	Female (%)
Smoking
Cooper et al. (2015)	fMRI Brain activity behaviour change	Longitudinal	Animated banner advertisements	50 All smokers	32.06 (12.61)^a 19–64 years^a	41^a
Chua et al. (2009)	fMRI Message tailoring effects	Cross-sectional	Written and audio (simultaneously) stimuli	24 All smokers	40.00 (11.20) Not provided	50
Chua et al. (2011)	fMRI Message tailoring effects	Longitudinal	Written and audio (simultaneously) statements	91 All smokers	37.50 (11.50) 21–55 years	48
Dinh-Williams et al. (2014)	fMRI Executive–affective connectivity	Cross-sectional	International Affective Picture System images	30 All smokers	31.80 (9.20) 18–55 years	50
Falk et al. (2011)	fMRI Brain activity behaviour change	Longitudinal	Video PSAs	31 All smokers	45.00 (10.10) 28–69 years	48
Kessels et al. (2010)	ERP Defensive processing of self-relevant health information	Cross-sectional	Images	30 15 smokers; 15 non-smokers	21.00 (1.50) 18–25 years	48
Kessels et al. (2014)	ERP Defensive processing of self-relevant health information	Cross-sectional	Video PSAs	22 All smokers	20.06 18–22 years	100
Kessels et al. (2014)	ERP Defensive processing of self-relevant health information	Cross-sectional	Video PSAs	25 10 smokers; 12 non-smokers	22.62 (3.56)^a 18–33 years^a	68^a
Langleben et al. (2009)	fMRI Message sensation value	Cross-sectional	Video PSAs	18 All smokers	23.00 (7.00) 18–48 years	17
Wang et al. (2013)	fMRI Message sensation value and argument strength	Longitudinal	Video PSAs	71 All smokers	30.21 (9.69) 18–49 years	48
Nutrition
Kessels et al. (2011)	ERP Message tailoring	Cross-sectional	Written stimuli	41	Low-threat condition: 21.30 (2.54) High-threat condition: 22.14 (2.57) 19–28 years	100
Ruiter et al. (2006)	ERP Message tailoring	Cross-sectional	Written stimuli	29	21.38 (2.15) 17–26 years	69
Sun safety
Falk et al. (2010)	fMRI Brain activity behaviour change	Longitudinal	Text/image slides; audio via earphones	20	22.80 (3.60) Not provided	50
Physical activity
Falk et al. (2015)	fMRI Self-affirmation on message processing and behaviour change	Longitudinal	Visual/ written slides	67	33.42 (13.04) Not provided	61
Narcotic substance/antidrug
Ramsay et al. (2013)	fMRI Co-activity of affective and executive networks in message processing	Cross-sectional	Video PSAs	70	16.75 (1.54) 15–19 years	50
Weber et al. (2015)^b	fMRI Message sensation value and argument strength	Cross-sectional	Video PSAs	28 14 high-risk; 14 low-risk cannabis users	20.30 (1.50) 18–25 years	100
Safe sex
Seelig et al. (2014)	fMRI Message sensation value	Longitudinal	Video PSAs	39	23.40 (2.90) 19–31 years^a	50^a
Other
Falk et al. (2009)	fMRI Cultural differences and message processing	Cross-sectional	Text-based phrases	15	20.75 (3.21) Not provided	47
Falk et al. (2009)	fMRI Cultural differences and message processing	Cross-sectional	Text-based phrases	14	22.06 (3.96) Not provided	79
Falk et al. (2009)	fMRI Cultural differences and message processing	Cross-sectional	Video PSAs	27	20.11 (2.66) Not provided	56

SD: standard deviation; fMRI: functional magnetic resonance imaging; ERP: event-related potential; PSA: public service announcement.

N = total number of participants prior to exclusion.

Sample statistics were only provided for participants who were not excluded.

Demographic information based only on those participants who undertook the neuroimaging component of research.

ERP studies

Five studies used ERPs to assess attention and/or processing towards health communication stimuli (see Kessels et al., 2010, 2011, 2014; Ruiter et al., 2006). Two of these studies focused on message tailoring (Kessels et al., 2011; Ruiter et al., 2006) and three studies examined defensive processing of self-relevant health information (Kessels et al., 2010, 2014).

Message tailoring

Kessels et al. (2011) and Ruiter et al. (2006) devised auditory dual-paradigm tasks whereby individuals were exposed to two auditory tones while reading either a tailored (i.e. high personal relevance) or non-tailored (i.e. low personal relevance) education campaign messages about nutrition.⁵ Specifically, participants were required to respond to the high-frequency tones (occurrence rate of 17%) by pressing a button; no response was required on presentation of the low-frequency tones (occurrence rate of 83%). Thus, as a result of responding to auditory cues, participants allocated more attention towards these cues, leaving less attentional resources available for the message. Furthermore, Kessels et al. (2011) included an additional between-group variable of threat (high vs. low threat associated with the negative consequences resulting from unhealthy eating). Examining early attentional processes, Ruiter et al. (2006) reported that individuals exposed to a tailored message demonstrated a larger N100 mean amplitude along the frontal–central midline on presentation of the auditory cues, suggesting less rather than greater early attention towards that message, compared to those individuals exposed to the non-tailored message. In other words, participants allocated greater attentional resources towards the auditory cues (a more pronounced N100) and, in turn, less attentional resources were allocated towards the message. No significant N100 effects involving message tailoring or threat were reported in the study by Kessels et al. (2011). Given that the studies had a number of similarities including dual-task paradigms, written message stimuli and recruitment of participants from the same population (undergraduate students), the mixed findings suggest that early attention processes towards health communication messages warrant further investigation. However, despite these inconsistent findings, both studies found that the tailored messages were processed to a greater extent, as assessed by the P300 amplitudes, compared to the non-tailored messages.

Kessels et al. (2011) found that individuals elicited larger P300 peak amplitudes towards auditory cues on presentation of high-threat compared to low-threat messages, indicating greater processing of the low-threat than high-threat messages. However, in contrast to these ERP results, in the accompanying self-report ratings, participants had reported giving similar levels of attention towards each of the four message conditions. Subsequent analyses confirmed that there were no significant differences in self-reported attention ratings between the message conditions (tailored and non-tailored, high- and low-threat messages). These findings highlight the discrepancies that may exist between self-report and neurocognitive measures of persuasive effects and, in particular, emphasize the insights which may be offered via the increased sensitivity of objective measures.

Defensive processing

Kessels et al. (2010) devised a cueing task which involved participants processing high and low physical threat smoking-related images in two trial types: valid trials, where subsequent cues were presented in the same position as the image (assessment of attention capturing processes), and invalid trials, where subsequent cues were presented in a different position than the previous image (assessment of attention disengagement). For the valid trials, the findings revealed that regardless of smoking status (smoker vs. non-smoker), participants showed greater processing (as assessed by larger P300 amplitudes) towards the high-threat than low-threat images. However, on the invalid trials, smokers showed more effective disengagement from the high-threat images when the cue was presented in the opposite field, suggesting greater defensive processing. These findings highlight that self-relevant threatening health information messages that are designed to target those individuals most at risk (e.g. smokers compared to non-smokers) may lead to attentional disengagement and, as such, be less persuasive than other types of health communication approaches (e.g. low-threat approaches).

Providing some support for the invalid trial findings of Kessels et al. (2010), Kessels et al. (2014) found that smokers elicited larger P300 responses on presentation of auditory cues presented in high-threat compared to low-threat anti-smoking advertisements (Studies 1 and 2), reflecting less processing for high-threat messages (since greater attentional resources were dedicated towards the auditory cue task). Non-smokers, however, elicited larger P300 amplitudes on presentation of auditory cues in the low-threat compared to high-threat advertisements (Study 2, Session 1). These findings are consistent with previous self-report research which has reported that self-relevant threatening health information may lead to defensive and/or avoidance responses (Witte and Allen, 2000).

fMRI studies

Of the 15 fMRI studies, 2 examined message tailoring effects (Chua et al., 2009, 2011), 2 investigated the influence of message sensation value alone (Langleben et al., 2009; Seelig et al., 2014), 2 examined the interaction between message sensation value and argument strength (Wang et al., 2013; Weber et al., 2015), 3 examined cultural differences and message persuasiveness (Falk et al., 2009), 4 assessed whether neural activity could explain additional variance in behaviour change unaccounted for by self-report measures of persuasion (Cooper et al., 2015; Falk et al., 2010, 2011, 2015) and 2 studies examined executive–affective connectivity (Dinh-Williams et al., 2014; Ramsay et al., 2013).

Message tailoring

Two studies examined tailoring effects of smoking cessation messages (i.e. Chua et al., 2009, 2011). In both studies, participants showed significantly greater neural activity in areas of the medial prefrontal cortex (MPFC), precuneus and posterior cingulate cortex (regions reported to reflect self-related processing; e.g. Northoff and Bermpohl, 2004; Northoff et al., 2006) when exposed to the high-tailored compared to the low-tailored messages. Furthermore, Chua et al. (2011) reported that greater activity in the dorsomedial prefrontal cortex (dMPFC) towards the tailored compared to the neutral messages also significantly predicted subsequent smoking cessation 4 months after initial message exposure.

Two studies were identified which examined message sensation value and an additional two studies examined the interaction between message sensation value and argument strength, all of which exposed participants to video-based television advertisements. These studies can be further classified and discussed in accordance with two theories of persuasion, the limited capacity model of motivated mediated message processing ((LC4MP); Langleben et al., 2009; Seelig et al., 2014) and the elaboration likelihood model ((ELM); Wang et al., 2013; Weber et al., 2015).

LC4MP

The LC4MP is an information-processing model which involves three sub-processes (i.e. encoding, storage and retrieval; Lang, 2000). The model postulates that individuals have a limited number of cognitive resources which can be used to process messages. Furthermore, the model predicts that other aspects of the individual (e.g. motivation/ emotional processes) and message (e.g. message sensation value) affect how, and to what extent, the message is processed. Langleben et al. (2009) and Seelig et al. (2014) examined smoking behaviour and safe sex (condom use), respectively. In both of these studies, messages comprised high- and low-arousing/stimulating messages, and participants displayed greater activity in areas of the prefrontal and temporal cortex on presentation of the low message sensation value advertisements compared to high message sensation value advertisements. These neural regions have been previously associated with a number of cognitive processes, such as decision-related and memory processes (Reckless et al., 2014; Wagner et al., 1998), and support the LC4MP hypothesis that message characteristics influence message processing. In contrast, for the high (compared to low) message sensation value advertisements, Langleben et al. (2009) and Seelig et al. (2014) found that participants displayed greater activity in the occipital cortex, an area associated with visual processing (e.g. Grill-Spector et al., 1999). Furthermore, Seelig and colleagues’ self-report data, which examined memory recognition (presented immediately, and 3 weeks, after message exposure), revealed that participants were significantly more likely to remember the low message sensation value advertisements compared to the high sensation advertisements, providing some support for the LC4MP. Collectively, these findings reveal that message sensation value could affect the persuasiveness of health communication messages and that this effect may be evident across different types of behaviour, although the precise neural network mechanisms underlying this effect are still unclear.

Seelig et al. (2014) interpreted their findings as supporting their occipital–temporal disruption hypothesis to explain the poorer recall associated with high sensation messages. Specifically, they proposed that overstimulation of the occipital cortex by high sensation advertisements reduces the processing resources available for encoding of the material in temporal regions. However, their analysis of the interaction between activity in these regions failed to reach conventional statistical significance (with p < .06) and revealed a general pattern of negative interaction for both types of messages (albeit showing a trend towards weaker levels of this interaction for the low message sensation value messages). Furthermore, the impact of these interaction differences on recall is unknown.

ELM

The ELM proposes two pathways to information processing: the central pathway, used when there is a high degree of elaboration (i.e. greater processing), and the peripheral pathway, used when there is a low degree of elaboration (Petty and Cacioppo, 1986). When health communication messages are processed via the central pathway, factors related to the message content (e.g. individual’s knowledge of the risks associated with the behaviour) influence message processing. Alternatively, when messages are processed via the peripheral pathway, factors unrelated to the message (e.g. simple cues, such as source characteristics) have a stronger influence on message processing. It is important to note, however, that elaboration consists of a continuum and individuals are likely to use a combination of both the pathways to process message content.

The findings of Wang et al. (2013) and Weber et al. (2015) can be interpreted in accordance with the ELM. Wang et al. (2013) exposed participants to a series of anti-smoking advertisements that differed in message sensation value (high and low; within-group variables) and argument strength (high vs. low; between-group variables). The findings revealed that areas located in the occipitoparietal neural regions, specifically the inferior parietal lobe (IPL) and the left fusiform gyrus (FG), were more active on presentation of high message sensation value/high argument advertisements compared to the remaining three message conditions. While Langleben et al.’s (2009) and Seelig et al.’s (2014) findings revealed that message sensation value was a key component of neural activity in the occipital cortex, Wang et al.’s (2013) findings suggest that argument strength may also influence activity in this brain region. In accordance with the ELM, Wang et al.’s (2013) findings highlight that both message (i.e. argument strength and message sensation value) and individual (i.e. smoking status) characteristics influence message processing and subsequent behaviour change. Furthermore, and consistent with Chua et al. (2011), Wang et al. (2013) found that greater activity in the dMPFC on presentation of the advertisements predicted significantly lower cotinine levels (less smoking) 1 month after message exposure. Objective measures of message processing (such as fMRI), therefore, provided insight into subsequent behaviour change.

Weber et al. (2015) exposed participants to anti-drug advertisements that differed in message sensation value (high and low; within-group variables) and argument strength (high and low; within-group variables). Participants were also exposed to an additional eight control advertisements. The interaction of message sensation value and argument strength was observed in greater neural activity for the high-risk group (i.e. participants more susceptible to cannabis use) than the low-risk group in the following brain regions: superior temporal gyrus (STG), precuneus, frontal pole (FP) and middle frontal gyrus (MFG).⁶ These regions have been associated with language processing, self-referencing, decision-making and cognition processes (e.g. sematic processing), respectively (Bechara et al., 2000; Davis and Johnsrude, 2003; Laufer et al., 2011). Similar to Wang et al. (2013), the findings provide support for the ELM suggesting that higher order cognitive processes influence message processing and message persuasion. To further explore outcomes of persuasion, Weber et al. (2015) applied both self-report and fMRI data to predict self-reported perceived message effectiveness in two additional, larger datasets (N = 599 and N = 601). Their findings revealed that the STG and MFG were predictors of perceived message effectiveness in these independent datasets and, as such, they suggested that neural predictors could offer additional insight into self-reported persuasive effects beyond that offered via self-report measures alone. Furthermore, these findings highlight how neurocognitive studies, which generally comprise smaller sample sizes compared to the larger sample sizes typically needed in self-report research, can be used to predict larger population effects (see Berkman and Falk, 2013; Falk et al., 2012 for an overview of applying neurological measures to examine population-based message effects).

Cultural differences and message persuasiveness

Falk et al. (2009) conducted three studies to examine neural activity towards a series of text-based persuasive and unpersuasive message phrases⁷ that focused on a range of activities including, flossing and blood donation (Studies 1 and 2) and video-based advertisements (Study 3). In Study 1, young European-American adults viewed 20 blocks of five phrases, prior to completing self-report items that assessed message persuasiveness and argument type (i.e. rated if the phrases were information-based or feeling-based). In Study 2, young Korean adults undertook the same experiment, but with the text-based phrases in Korean rather than English. Study 3 consisted of 27 young European-American adults and examined whether similar neural regions were activated on presentation of a series of video-based commercials. The findings for Studies 1 and 2 revealed that the same brain regions (i.e. the left dMPFC, bilateral posterior superior temporal sulcus (pSTS), bilateral temporal poles (TPs), left ventrolateral prefrontal cortex (VLPFC), hippocampus and left lingual gyrus) were activated for both the European-American and Korean participants on presentation of the persuasive phrases compared to the unpersuasive phrases. Similar brain regions were activated on presentation of the video-based commercials in Study 3. These findings thus suggest that specific neural regions (i.e. the dMPFC, bilateral pSTS, bilateral TP and left VLPFC; areas associated with social cognition and affective processing) may be associated with the processing of persuasive messages across cultures and message types. However, cross-cultural differences were noted in the strength of neural activations across specific regions. For instance, while the middle and inferior occipital gyrus regions were more active among the Korean participants on presentation of the persuasive compared to unpersuasive phrases, brain regions associated with emotional and cognitive processes, such as the medial temporal lobe, posterior cingulate and ventral striatum, were more active among the European-American participants (see Falk et al., 2009 for the full list of brain regions). Further research is needed (e.g. using a bilingual sample with a within-group design for the language variable) to clarify whether this finding may reflect differences in the strength of association of the words used (e.g. familiarity in everyday usage) for the American-English message compared to the translated Korean message, or broader cultural differences.

Neural activity and behaviour change

Four studies examined whether neural activity, recorded with fMRI during stimulus exposure, explained additional variance in behaviour change beyond that accounted for by self-report measures of persuasion (Cooper et al., 2015; Falk et al., 2010, 2011, 2015). Falk et al. (2010) assessed individuals’ attitudes and intentions towards sunscreen use pre and post their exposure to a series of sun safety slides, with self-reported sunscreen use assessed 1 week after message exposure. Activation in one of the two⁸ regions of interest (ROIs), the MPFC, predicted increased sunscreen use when controlling for variance explained by the self-report measures of persuasion. Falk et al. (2011), in turn, identified that the ventral subregion of the MPFC was associated with subsequent smoking cessation when controlling for variance explained by the self-report measures of intentions, self-efficacy and self-relevance, 1 month after exposure to anti-smoking video-based television advertisements. Specifically, activity in the ventral subregion of the MPFC accounted for an additional 20.4 per cent in smoking cessation (assessed via carbon monoxide) over and above that which was explained by self-report measures of persuasion, the latter which accounted for 14.6 per cent of the variance in smoking cessation.

Cooper et al. (2015) examined three sub-regions of the MPFC, all which had been reported to reflect both self-related processing and value-related processing (i.e. personal, perceived value of a specific stimuli). These sub-regions of the MPFC predicted significant behaviour change in reduced cigarette use over and above self-reported intentions. Here, a follow-up phone call was used to examine post smoking behaviour (daily number of cigarettes smoked), approximately 40 days after advertisement exposure. Additionally, the findings revealed no significant relationships between self-reported measures of persuasion (intentions and self-efficacy) and MPFC activity, further highlighting the added benefits of incorporating neural activity to examine behaviour change.

Falk et al. (2015) examined the influence of self-affirmation on participant responses to 50 physical activity message slides and behaviour change in terms of reducing sedentary behaviour at 1 month post-message exposure.⁹ Accelerometers assessed both pre-message (1 week prior to message exposure) and post-message sedentary behaviour. Participants allocated to the self-affirmation condition (provided with additional time to think about the information presented on the slides, in relation to their highest core values¹⁰ based on a prior self-report measure) compared to the participants in the control condition (viewed situations related to their lowest core values) showed greater activity in the ventromedial prefrontal cortex (VMPFC) region on presentation of the slides and increased physical activity 1 month after message exposure. In line with Cooper et al. (2015) and Falk et al. (2010, 2011), neural activity in the VMPFC accounted for additional variance in behaviour change over and above that accounted for by self-reported measures of attitudes towards physical activity. Although it is acknowledged that further research is required to examine the neurological processes underlying message processing, collectively the fMRI findings from Cooper et al. (2015), Chua et al. (2011), Falk et al. (2010, 2011, 2015), Wang et al. (2013) and Weber et al. (2015) suggest that neural activity could be used to assess message processing and/or used as a predictor of behaviour change.

Executive–affective connectivity

Ramsay et al. (2013) examined the extent to which activity in the executive neural networks (e.g. lateral prefrontal cortex) interacted with activity in the socioemotional networks (e.g. amygdala). Participants were exposed to 10 strong and 10 weak argument strength anti-drug video-based advertisements and 10 non-drug advertisements. After viewing the advertisements, participants self-reported the extent to which they perceived the advertisements as convincing and arousing. Compared to the non-drug advertisements, when participants viewed the weak and strong anti-drug advertisements, they displayed greater arousal-related activity in areas of the socioemotional networks including bilateral amygdala, medial orbital frontal cortex (MOFC), paracingulate gryus, bilateral hippocampus and STG. For the anti-drug advertisements, the findings revealed that for the strong (compared to the weak) advertisements, greater arousal-related activity was found in executive neural areas such as the bilateral MFG and the left inferior frontal gyrus (IFG). To further examine connectivity between the executive and socioemotional networks, the authors selected two seed regions, namely, the left IFG and MFG, to assess the extent to which these areas co-activated on presentation of the persuasive advertisements, independent of arousal-related activity. While activity in the left IFG revealed statistically significant predictions, activity in the MFG did not show significant connectivity. Specifically, the left IFG showed significantly more positive co-activity in socioemotional regions, including the amygdala and insula, on presentation of the strong compared to weak advertisements. Overall, these findings revealed that both socioemotional and executive neural networks may be involved in message processing.

Dinh-Williams et al. (2014) examined executive–affective connectivity in adult smokers exposed to series of aversive (unpleasant/arousing) smoking- and non-smoking-related images and neutral images. Compared to the non-smoking images, there was significant negative connectivity between the left IFG and areas linked to self-relevant message processing, namely, the MPFC and precuneus, as well as the right anterior cingulate cortex and insula during presentation of the smoking-related images. This finding is similar to Ramsay et al. (2013) who found that the left IFG interacted with socioemotional networks (amygdala and insula) during message processing. The strength of this relationship between the aforementioned neural regions in Dinh-Williams et al. (2014) was strongest for those participants who had started smoking at a younger age. As proposed by Dinh-Williams et al. (2014), these findings suggest that greater activity in the left IFG may interfere with activity occurring in brain regions linked to self-relevant message processing (e.g. the MPFC) and, in turn, act to reduce the processing of self-relevant aversive health images. Thus, those individuals most at risk may be less likely to be persuaded by images that reflect the negative effects of specific health behaviours.

Discussion

This article systematically reviewed the scientific literature that has applied neurocognitive techniques to examine message processing and their contribution to/prediction of message persuasion outcomes with particular focus on outcomes relating to behaviour change. Overall, evidence from the reviewed studies suggests that neurocognitive measures (ERP and fMRI) could be used alongside self-report measures to enhance understanding of both persuasive processing and outcomes with respect to health messages.

Message tailoring

The ERP and fMRI findings revealed that cognitive processing (as assessed by the P300 amplitude) and self-related processing (as assessed by greater activity in areas of the MPFC, precuneus and posterior cingulate) may be more prominent in the persuasive processing of self-relevant health advertisements. Although these findings were consistent with those reported in previous self-report data from message tailoring research (e.g. Skov-Ettrup et al., 2014), one study in the current review found that self-report measures of perceived attention given to a message were inconsistent with the ERP findings (Kessels et al., 2011). Thus, had conclusions been drawn only from the self-report data in Kessels et al. (2011), it may have been concluded that the tailored and non-tailored messages did not affect attentional processes. Incorporating a range of both self-report and neurocognitive measures may not only provide a more in-depth understanding of message processing but also provide greater insights into the limitations and capabilities of traditional self-report measures.

Message sensation value and threat

The ERP findings revealed that self-relevant advertisements that contained high-threat stimuli could potentially lead to attentional disengagement. The fMRI data showed that greater activity in neural regions thought to reflect higher order cognitive processes was elicited on presentation of low message sensation value advertisements compared to high message sensation value advertisements which, in turn, resulted in greater visual processing. Health messages that focus on risky behaviours have typically used threat-based communication appeals to target those individuals most at risk (e.g. Hastings et al., 2004; Lewis et al., 2007). However, theoretical models (i.e. the protection motivation theory; Rogers, 1983, and the extended parallel process model; Witte, 1992) and empirical research (e.g. Kessels and Ruiter, 2012) have identified that complementing threatening health information with coping information may enhance perceptions of coping, comprising response efficacy and self-efficacy. More recently, empirical evidence has also reported that other messages’ approaches, such as positive-based emotional appeals, may be an alternative option to persuade individuals most at risk (see: Lewis et al., 2007). Taken together, the ERP and fMRI findings reported in this review may suggest that alternative message approaches should be explored for their potential differential effects on message processing and persuasion.

Brain-as-predictor approach

Many of the fMRI studies examined actual behaviour change via objective measures and highlighted that activity in specific brain regions could be used as predictors of later behaviour change. More specifically, Cooper et al. (2015), Chua et al. (2011), Falk et al. (2011, 2015) and Wang et al. (2013) found that activity in the MPFC during message exposure predicted later behaviour change (reduction in smoking behaviour and sedentary activity, increased sunscreen use), with follow-up periods ranging from 1 week to 4 months. Furthermore, activity in the MPFC explained additional variance in behaviour change beyond that accounted for by self-reported measures of message acceptance (i.e. attention, intentions and self-efficacy). It should be noted, however, that two of the five studies which examined behaviour change in smoking behaviour recruited participants who were either enrolled in a quit smoking programme or were interested in quitting smoking and thus were already motivated to change their behaviour. However, despite pre-existing motivations for behaviour change, the findings suggest that activity in the MPFC may be a predictor of later behaviour change and could potentially add to the existing knowledge that has been acquired through self-report measures of persuasive outcomes.

Weber et al. (2015) demonstrated that neural activity in a smaller fMRI study could be used as a predictor of message effectiveness in larger population samples. This finding is consistent with previous research which examined neural activity in a smaller sample of 31 smokers and found that activity in the MPFC (ROI) could predict the success of anti-smoking campaigns at the population level (Falk et al., 2012). Falk et al. (2012) also reported that self-report measures of message effectiveness did not predict advertisement success. Drawing upon these research findings, campaign designers could potentially incorporate neurocognitive measures to examine the persuasiveness of new advertisements on a smaller group of individuals, prior to refining and subsequently releasing the advertisements to the general population, to enhance the success of health communication campaigns.

Implications of these findings for theories of persuasion

The body of findings reviewed herein is of sufficient strength and consistency to warrant a fresh look at the Falk et al.’s (2009) so-called ‘persuasion network’. Falk et al. (2009) proposed a neurocognitive network associated with persuasion, with neural areas including the left dMPFC, bilateral pSTS, bilateral TP and left VLPFC, areas previously implicated in social cognition, mentalizing, memory selection and emotional reappraisal. More recent research reviewed herein supports the proposed neurocognitive network, finding that specific neural areas such as the dMPFC are associated with persuasion (Chua et al., 2009, 2011; Wang et al., 2013). Furthermore, Wang et al. (2013) found greater activity in the occipitoparietal region on presentation of high sensation seeking/high argument advertisements, while Weber et al. (2015) showed that individuals who were more susceptible to the risk portrayed in the advertisements had greater activity in the STG, FP and MFG, areas aligned with brain regions initially proposed a priori by Falk et al. (2009) to also be important for persuasion (e.g. self-referential processing).

Ramsay et al. (2013) and Dinh-Williams et al. (2014) lend further support and extension to Falk et al.’s (2009) persuasion network by highlighting that executive and socioemotional networks interact when individuals are exposed to persuasive arguments. Specifically, co-activation of the left IFG and amygdala was increased, suggesting greater message processing, on presentation of the strong (vs. weak) anti-drug messages in Ramsay et al. (2013). Dinh-Williams et al. (2014), however, found that negative connectivity between the left IFG and insula was greater for drug compared to non-drug messages, suggesting that for drug users, targeted aversive messages were less persuasive. Collectively, these findings highlight that by examining executive and socioemotional networks simultaneously, researchers may gain a greater understanding on how these networks interact to increase (or decrease) message persuasion for different risky behaviours.

With increasing evidence for the utility of neurocognitive measures in elucidating brain-based mechanisms of persuasive processes, models such as Falk et al.’s (2009) so-called ‘persuasion network’ offer a unique perspective supported by objective neurological data to the previously dominant theories of message processing and persuasion such as the TPB, ELM and LC4MP. Studying the brain in this context offers the potential to not only overcome limitations associated with self-report measures but also examine and enhance/develop theories of message processing and persuasion focusing on neurocognitive mechanisms. Furthermore, given that health communication studies have reported that measures of brain activity can account for additional variance in behaviour change above and beyond self-report measures of attitudes and intentions (e.g. Falk et al., 2011), it is important that future research continues applying neurocognitive measures to further enhance our understanding of the underlying components of message processing and subsequent persuasion.

Self-report versus neurocognitive measures

Self-report measures are often employed to examine message processing and/or message acceptance. Self-report measures enable researchers to collect data from large sample sets and are a cheaper alternative to neurocognitive measures. However, these measures are susceptible to participant bias effects (e.g. social desirability) and are unable to assess unconscious processes, such as moment-by-moment changes in attention and/or cognitive processing. Alternatively, neurocognitive measures are less susceptible to participant biases compared to self-report measures as they assess automatic/uncontrolled processes. However, it is acknowledged that such neurocognitive measures are more expensive (e.g. cost of the equipment/materials, as well as greater collection time and expertise required of the administrator) and may be less accessible. As previously discussed, ERPs are a high-temporal resolution method which are sensitive to the changes in attention and cognitive processing (Duncan et al., 2009). The basic equipment and software required to collect ERP data is also significantly cheaper and therefore more accessible to research institutions, compared to fMRI. fMRI, however, has low temporal resolution due to slow blood flow response but has high spatial resolution which is more suited for source localization (i.e. identify neural regions and structures involved in processing; Logothetis, 2008). As demonstrated in this review, both ERPs and fMRI are feasible measures of attention and/or cognitive processing; however, fMRI may be more sensitive when examining neural regions as predictors of message effectiveness and/or behaviour change. Given that the current review has highlighted that persuasion is a socioemotional process, involving cortical (e.g. prefrontal cortex) and subcortical structures (e.g. subcortical limbic structures), a strength of fMRI is its ability to assess activity in both types of regions. Overall, and as previously argued by others (e.g. Berkman and Falk, 2013), a multi-method approach using neurocognitive measures in conjunction with existing measures of self-report may provide a more comprehensive understanding of processing and persuasive outcomes of health communication messages.

Limitations

The review has several limitations that must be acknowledged. First, this review was restricted to articles that were published in PsycINFO, PubMed, Web of Science and ProQuest Psychology Journals. Unpublished studies and those published outside these databases were not considered for inclusion, and therefore, the conclusions drawn from this review may be subject to publication bias. Second, three studies included in this review contained a mix of messages, some of which were pro-social/altruistic behaviours (e.g. blood donation) and therefore were not strictly health-related behaviours (Falk et al., 2009). Thus, the results from these studies may have been influenced by the other objects and activities examined in Falk et al. (2009). Finally, only 20 studies met the inclusion requirements, and thus conclusions were based on a limited number of studies. Given that examining neurocognitive activity in response to health messages is a rapidly growing area of interest, future research is required to further examine the utility of incorporating such measures in order to further understand the persuasive process and to test the persuasiveness of new messages prior to refining and releasing to the public. Finally, this review focused upon articles that dealt with persuasive messages in health contexts, rather than persuasion more generally. It is as yet unknown how the process of persuasion may differ, particularly using neurological measures, in health communication contexts compared with say, brand and product advertising, although this is an interesting avenue to explore in future research.

Conclusion

Various theoretical frameworks, such as the TPB, have often been applied to explain behaviour change. The TPB proposes that attitudes, subjective norms and perceived behaviour control predict intentions which, in turn, predict behaviour change (Ajzen, 1991). However, despite evidence suggesting that these variables influence behaviour (e.g. Fortier et al., 2009), there still remains considerable variance in behaviour change which is unaccounted for by the aforementioned constructs. The current review revealed that brain activity, specifically MPFC activity, a region reported to reflect self-related processing, can account for additional variance in behaviour change above and beyond self-report measures of attitudes and intentions. Identifying additional factors involved in message processing via neurocognitive measures (e.g. self- and value-related processing and self-affirmation, as identified from this review) could be used to further enhance the existing theories of behaviour change and consequently provide additional insights into the processes involved in health behaviour change and persuasion.

Overall, evidences from the studies presented herein indicate that neurocognitive measures can be applied successfully to examine the persuasiveness of health communication messages across different behaviours, message types, populations and cultural groups. A multi-method approach using neurocognitive measures with self-report measures may enable a more comprehensive understanding of the persuasion process and outcomes of health communication messages. Furthermore, in the health communication context, there is an opportunity to not only extend upon the research presented herein which examined messages designed to target risk-taking activities and promote healthier behaviours but also apply neurocognitive techniques to assess processing of messages that encourage individuals to be proactive in detecting potential life-threatening illness (e.g. cancer screening advertisements). Applying a multi-method approach to examine the persuasive process may lead to improvements in message design and, ultimately, healthier and safer behaviours in high-risk groups of individuals.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

Notes

References

Ajzen

(1991) The theory of planned behavior. Organizational Behavior and Human Decision Processes 50: 179–211. DOI: 10.1016/0749-5978(91)90020-T.

Bechara

Damasio

(2000) Emotion, decision making and the orbitofrontal cortex. Cerebral Cortex 10(3): 295–307. DOI: 10.1093/cercor/10.3.295.

Berkman

Falk

(2013) Beyond brain mapping: Using neural measures to predict real-world outcomes. Current Directions in Psychological Science 22(1): 45–50. DOI: 10.1177/0963721412469394.

Chua

Jasinska

et al . (2011) Self-related neural responses to tailored smoking-cessation message predicts quitting. Nature Neuroscience 14(4): 426–427. DOI: 10.1038/nn.2761.

Chua

Liberzon

Welsh

et al . (2009) Neural correlates of message tailoring and self-relatedness in smoking cessation programming. Biological Psychiatry 65: 165–168. DOI: 10.1016/j.biopsych.2008.08.030.

Cooper

Tompson

O’Donnell

et al . (2015) Brain activity in self-and value-related regions in response to online antismoking messages predicts behaviour change. Journal of Media Psychology.

Coull

(1998) Neural correlates of attention and arousal: Insights from electrophysiological, functional neuroimaging and psychopharmacology. Progress in Neurobiology 55: 343–361. DOI: 10.1016/S0301-0082(98)00011-2.

Cui

Bray

Bryant

et al . (2011) A quantitative comparison of NIRS and fMRI across multiple cognitive tasks. NeuroImage 54: 2808–2821. DOI: 10.1016/j.neuroimage.2010.10.069.

Davis

Johnsrude

(2003) Hierarchical processing in spoken language comprehension. Journal of Neuroscience 23(8): 3423–3431.

10.

Dillard

Shen

Vail

(2007) Does perceived message effectiveness cause persuasion or vice versa? 17 consistent answers. Human Communication Research 33: 467–488. DOI: 10.1111/j.1468-2958.2007.00308.x.

11.

Dinh-Williams

Mendrek

Dumais

et al . (2014) Executive-affective connectivity in smokers viewing anti-smoking images: An fMRI study. Psychiatry Research: Neuroimaging 224: 262–268. DOI: 10.1016/j.pscychresns.2014.10.018.

12.

Duncan

Barry

Connolly

et al . (2009) Event-related potentials in clinical research: Guidelines for eliciting, recording, and quantifying mismatch negativity, P300, and N400. Clinical Neurophysiology 120: 1883–1908. DOI: 10.1016/j.clinph.2009.07.045.

13.

Falk

Berkman

Lieberman

(2012) From neural responses to population behavior: Neural focus group predicts population-level media effects. Psychological Science 23(5): 439–445. DOI: 10.1177/0956797611434964.

14.

Falk

Berkman

Mann

et al . (2010) Predicting persuasion-induced behavior change from the brain. Journal of Neuroscience 30(25): 8421–8424. DOI: 10.1523/JNEUROSCI.0063-10.2010.

15.

Falk

Berkman

Whalen

et al . (2011) Neural activity during health messaging predict reductions in smoking above and beyond self-report. Health Psychology 30(2): 177–185. DOI: 10.1037/a0022259.

16.

Falk

O’Donnell

Cascio

et al . (2015) Self-affirmation alters the brain’s response to health messages and subsequent behavior change. Proc Natl Acad Sci USA 112(7): 1977–1982. DOI: 10.1073/pnas.1500247112.

17.

Falk

Rameson

Berkman

et al . (2009) The neural correlates of persuasion: A common network across cultures and media. Journal of Cognitive Neuroscience 22(10): 2447–2459. DOI: 10.1162/jocn.2009.21363.

18.

Fortier

Kowal

Lemyre

et al . (2009) Intentions and actual physical activity behaviour change in a community-based sample of middle-aged women: Contributions from the theory of planned behavior and self-determination theory. International Journal of Sport and Exercise Psychology 7: 46–67. DOI: 10.1080/1612197X.2009.9671892.

19.

Grill-Spector

Kushnir

Edelman

et al . (1999) Differential processing of objects under various viewing conditions in the human lateral occipital complex. Neuron 24: 187–203.

20.

Hastings

Stead

Webb

(2004) Fear appeals in social marketing: Strategic and ethical reasons for concern. Psychology and Marketing 21: 961–986. DOI: 10.1002/mar.20043.

21.

Kessels

Ruiter

(2012) Eye movement responses to health messages on cigarette packages. BMC Public Health 12: 352–362. DOI: 10.1186/1471-2458-12-352.

22.

Kessels

Ruiter

Jansma

(2010) Increased attention but more efficient disengagement: Neuroscientific evidence for defensive processing of threatening health information. Health Psychology 29(4): 346–354. DOI: 10.1037/a0019372.

23.

Kessels

Ruiter

Brug

et al . (2011) The effects of tailored and threatening nutrition information on message attention. Evidence from an event related potential study. Appetite 56(1): 32–38. DOI: 10.1016/j.appet.2010.11.139.

24.

Kessels

Ruiter

Wouters

et al . (2014) Neuroscientific evidence for defensive avoidance of fear appeals. International Journal of Psychology 49(2): 80–88. DOI: 10.1002/ijop.12036.

25.

Lang

(2000) The limited capacity model of mediated message processing. Journal of Communication 50(1): 46–70.

26.

Langleben

Loughead

Ruparel

et al . (2009) Reduced prefrontal and temporal processing and recall of high ‘sensation value’ ads. NeuroImage 46: 219–225. DOI: 10.1016/j.neuroimage.2008.12.062.

27.

Laufer

Negishi

Lacadie

et al . (2011) Dissociation between the activity of the right middle frontal gyrus and the middle temporal gyrus in processing semantic priming. PLoS ONE 6(8): e22368. DOI: 10.1371/journal.pone.0022368.

28.

Lewis

Watson

Tay

et al . (2007) The role of fear appeals in improving driver safety: A review of the effectiveness of fear-arousing (threat) appeals in road safety advertising. International Journal of Behavioral Consultation and Therapy 3: 203–222.

29.

Liberati

Altman

Tetzlaff

et al . (2009) The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: Explanation and elaboration. Annals of Internal Medicine 151(4): 65–94.

30.

Logothetis

(2008) What we can do and what we cannot do with fMRI. Nature 453: 869–878.

31.

Northoff

Bermpohl

(2004) Cortical midline structures and the self. Trends in Cognitive Sciences 8(3): 102–107. DOI: 10.1016/j.tics.2004.01.004.

32.

Northoff

Heinzel

de Greck

et al . (2006) Self-referential processing in our brain – A meta-analysis of imaging studies on the self. NeuroImage 31: 440–457. DOI: 10.1016/j.neuroimage.2005.12.002.

33.

Petty

Cacioppo

(1986) The elaboration likelihood model of persuasion. Advances in Experimental Social Psychology 19: 123–205. DOI: 10.1016/S0065-2601(08)60214-2.

34.

Polich

(2007) Updating P300: An integrative theory of P3a and P3b. Clinical Neurophysiology 118: 2128–2148.

35.

Ramsay

Yzer

Luciana

et al . (2013) Affective and executive network processing associated with persuasive antidrug messages. Journal of Cognitive Neuroscience 25(7): 1136–1147. DOI: 10.1162/jocn_a_00391.

36.

Reckless

Ousdal

Server

et al . (2014) The left inferior frontal gyrus is involved in adjusting response bias during a perceptual decision-making task. Brain and Behavior 4(3): 398–407. DOI: 10.1002/brb3.223.

37.

Rogers

(1983) Cognitive and physiological processes in fear appeals and attitude change: A revised theory of protection motivation. In: Cacioppo

Petty

(eds) Social Psychophysiological: A Source Book. London: Guildford Press, pp. 153–176.

38.

Ruiter

Kessels

Jansma

et al . (2006) Increased attention for computer-tailored health communications: An event-related potential study. Health Psychology 25(3): 300–306. DOI: 10.1037/0278-6133.25.3.300.

39.

Seelig

Wang

A-L

Jaganathan

et al . (2014) Low message sensation health promotion videos are better remembered and activate areas of the brain associated with memory encoding. PLoS ONE 9(11): e113256. DOI: 10.1371/journal.pone.0113256.

40.

Skov-Ettrup

Ringgaard

Dalum

et al . (2014) Comparing tailored and untailored text messages for smoking cessation: A randomized controlled trial among adolescent and young adult smokers. Health Education Research 29(2): 195–205. DOI: 10.1093/her/cyt112.

41.

Wagner

Schacter

Rotte

et al . (1998) Building memories: Remembering and forgetting of verbal experiences as predicted of brain activity. Science 281: 1188–1191.

42.

Wang

Ruparel

Loughead

et al . (2013) Content matters: Neuroimaging investigation of brain and behavioral impact of televised anti-tobacco public service announcements. Journal of Neuroscience 33(17): 7420–7427. DOI: 10.1523/JNEUROSCI.3840-12.2013.

43.

Weber

Huskey

Mangus

et al . (2015) Neural predictors of message effectiveness during counterarguing in antidrug campaigns. Communication Monographs 82(1): 4–30. DOI: 10.1080/03637751.2014.971414.

44.

Witte

(1992) Putting the fear back into fear appeals: The extended parallel process model. Communication Monographs 59: 329–349.

45.

Witte

Allen

(2000) A meta-analysis of fear appeals: Implications for effective public health campaigns. Health Education & Behavior 27: 591–615. DOI: 10.1177/109019810002700506.