Friday, June 17, 2022

More intelligent individuals reported lower symptom counts, less chronic conditions, lower rates of depression, less frequent visits to the doctor, and fewer limitations in their daily lives

An intelligent mind in a healthy body? Predicting health by cognitive ability in a large European sample. Jonathan Fries Jakob Pietschnig. Intelligence, Volume 93, July–August 2022, 101666.


• We demonstrate that cognitive ability predicts various aspects of health in adults over 55 years.

• Effect sizes are modest, but may have a considerable impact on population level.

• The most closely g-related construct (mathematical reasoning) predicted indicators of health most consistently.

• Environmental and behavioral risk factors do not play a meaningful role for the intelligence-health association.

Abstract: Intelligence has been consistently demonstrated to be a predictor of health outcomes. However, the exact mechanisms are subject of debate. Environmental and behavioral risk factors have been suggested to affect the intelligence-health association, but the available literature has mostly focused on children and young adults. Here, we aimed to investigate the intelligence-health association in older adults. We analyzed data from the Study of Health and Retirement in Europe (SHARE), a representative longitudinal survey in which participants above 50 years of age (N range = 10,000-30,000+) were interviewed in seven waves from 2004 to 2017. Indicators of physical and mental health (e.g., number of symptoms; self-reported depression) were associated with cognitive function variables (mathematical reasoning, word recall, verbal fluency) which were used as proxy measures for intelligence. Behavioral and environmental risk factors (e.g., legal drug consumption, physical inactivity, work environment) were examined as potential moderator variables for the intelligence-health association. More favorable health outcomes were modestly, but consistently associated with higher cognitive ability across variables (r range = |0.13|-|0.29|). Mixed-model Poisson regression analyses showed a reduction of 11% in self-reported symptom numbers with each unit increase in mathematical reasoning. Environmental and behavioral risk factors exhibited mostly trivial moderating effects on the intelligence-health association. Our findings reveal a positive association of intelligence and health in a representative longitudinal European sample. Environmental and behavioral risk factors offered little explanatory value for this association, suggesting a different underlying mechanism such as a general fitness factor that affects both intelligence and health.

4. Discussion

Here, we investigated the associations of cognitive ability and health in a representative longitudinal sample of EU residents above the age of 50. Specifically, we examined various indicators of physical and mental health, along with behavioral and environmental risk factors as well as cognitive functioning. In our analyses, a consistent pattern emerged: cognitive ability was associated with the rate of chronic illnesses, symptoms, limitations in daily activities, and other indicators of health. This was true both for individual cognitive ability indicators and the g-factor approximation we calculated from these indicators. Environmental and behavioral risk factors, such as smoking, alcohol consumption, BMI, or work environment showed little moderating effects. Physical inactivity was the only covariate that exhibited moderating effects on the intelligence-health association.

The strength of the associations was small to moderate in size. Such effects have practical meaning in the short term but can be considered even more meaningful on a larger scale and in the long term. Intelligence and health are relevant to every single human. Thus, even small differences cumulate over time and can have tremendous consequences not only on an individual but perhaps more importantly, a societal level (Funder & Ozer, 2019). Previous research has reported effect sizes of similar strength (for an overview, see Deary et al., 2021). The direction of effects was as expected: higher cognitive ability was positively related to more favorable health outcomes. More intelligent individuals reported lower symptom counts, less chronic conditions, lower rates of depression, less frequent visits to the doctor, and fewer limitations in their daily lives.

Notably, the largest effects were observed for self-perceived health. This variable encompasses a broad spectrum of health and may paint a more comprehensive picture than the other – more specific – indicators in the SHARE dataset. This makes sense because self-perceived or self-rated health is considered a reliable and robust predictor of health and mortality and has often been used in many studies on aging (e.g., French, Sargent-Cox, & Luszcz, 2012Machón, Vergara, Dorronsoro, Vrotsou, & Larrañaga, 2016). This can be attributed to the multi-dimensional and dynamic properties of this variable. Interviewees may not be aware of every detrimental condition from which they suffer, because some conditions may remain undiagnosed. However, self-perceived health assesses a wide range of sensations and symptoms that may indicate countless physical and mental health conditions in clinical and pre-clinical stages (Benyamini, 2011). Because self-perceived health arguably captures a larger proportion of variance in overall health than any other available variable, this variable may yield therefore the best representation of the true effect in terms of the intelligence-health association in the SHARE dataset.

It is important to consider the meaning of different directionalities of the observed effects; specifically, whether higher cognitive ability is the consequence of better health or whether higher cognitive ability leads to better health over the course of life. Though the longitudinal design of this study encompassed only a later part in participants' lives and does not allow for causal inferences, our results indicate that cognitive ability has an age-independent effect on health. If the intelligence-health association were a mere consequence of deteriorating health, this correlation would be expected to be substantially attenuated when controlling for age, because age is a robust predictor of self-perceived as well as objectively measured health (e.g., Cullati, Rousseaux, Gabadinho, Courvoisier, & Burton-Jeangros, 2014Rockwood, Song, & Mitnitski, 2011). However, this was not the case in our analyses, thus indicating that cognitive ability may be more likely to affect health instead of the other way round. Nonetheless, it seems plausible that good health also exerts positive effects on intelligence, thus representing a positive feedback loop in which higher cognitive ability facilitates better health which in turn helps maintaining high abilities.

Cancer on a general level did not correlate with any indicator of cognitive ability. This makes sense, because in contrast to many other health conditions (such as cardiovascular disease), a large number of cancer types does not depend on lifestyle factors and can therefore be expected not to be influenced by deliberated lifestyle choices. This observation is consistent with previous findings on this topic suggesting that lifestyle choices may not cause most types of cancer, with the notable exception of lung cancer (Calvin et al., 2017).

Across analyses, participants' numeracy scores proved to be the most robust predictor out of individual cognitive ability indicators: correlations decreased only to a minor degree when controlled for participant age, thus indicating that these associations were relatively unaffected by expectable age-related ability declines. In regression analyses, numeracy most reliably predicted health outcomes. This was expected because mathematical reasoning can be assumed to be more highly g-loaded than the other cognitive function indices in the SHARE dataset and is likely to estimate general cognitive ability more reliably (Peng et al., 2019). When controlled for age and education, the associations between cognitive ability and health were attenuated but remained meaningful and invariably maintained their respective directions which indicates the remarkable robustness of this effect. This was the case for correlations, as well as regression analyses.

Intelligence exhibited some associations with environmental and behavioral risk factors, but the directions of effects were not as clear-cut as with health. Higher cognitive ability was associated with lower rates of physical inactivity, which is consistent with previous findings (e.g., Wraw et al., 2018) that are suggestive of higher-intelligence individuals exhibiting higher ability and motivation to engage in vigorous physical activities (but see Kumpulainen et al., 2017, for contrasting results). Physical inactivity may be related to health literacy which is defined as the ability to gain access to information about health topics, as well as to interpret it and communicate about it. Health literacy is considered to be a prerequisite of informed health-related decision making (Berkman, Davis, & McCormack, 2010). Health literacy has been proposed as an important factor in the intelligence-health relationship because more intelligent individuals are assumed to obtain and process relevant health information more easily than less intelligent persons. Some research even suggests that health literacy is simply a context-specific component of general intelligence (Reeve & Basalik, 2014).

We had hypothesized BMI, smoking, and alcohol consumption to be negatively associated with cognitive ability. However, BMI was not meaningfully associated with cognitive ability, and neither was smoking. In previous accounts, smoking cessation has often been found to be linked with intelligence, but uptake of smoking has not (Taylor et al., 2003). Here, we only considered whether participants had ever smoked, but not if or when they quit which may have masked a potential association with cognitive ability.

Among risk factors, consumption of alcoholic beverages exhibited small, but meaningful positive correlations with cognitive ability, indicating that individuals with higher cognitive ability in fact consumed more alcohol than lower-ability persons. Previous studies have found similar correlations indicating more frequent overall alcohol consumption in subjects with higher childhood intelligence, but lower rates of problematic drinking behavior (Cheng & Furnham, 2013Kanazawa & Hellberg, 2010). These findings may be attributed to several different causes. It has been suggested that more intelligent individuals might be better equipped to avoid adverse health effects of drinking (e.g., by reducing their intake when they become aware about an onset of problematic drinking behavior). Considering that our analyses indicated no correlation with unfavorable health outcomes, our results are generally in line with this interpretation. Others have speculated that the success in certain (particularly white-collar) professions may depend to some extent on the willingness to drink alcohol in social settings that are typically related to more cognitively challenging jobs (Batty et al., 2008). Another possibility is that more intelligent individuals are better able to veil their problematic consumption from others or even physicians tend to misattribute problematic behaviors in more intelligent persons to less socially undesirable causes (Just-Ostergaard et al., 2019).

Unsurprisingly, participants who worked in higher-risk work environments scored lower on the cognitive function variables. This could indicate that low-risk, white-collar jobs are selected for via intelligence, which would be in line with previous research (Strenze, 2007). However, because in the present study cognitive ability was assessed at an average age of 64 years, one could argue that lower cognitive function may reflect overall poor health as a consequence of environmental factors such as work environment. Nevertheless, work environment exhibited only trivial bivariate associations with health which contrasts this interpretation.

Among environmental and behavioral risk factors, physical inactivity was the only one that showed consistent associations with health, especially regarding the number of limitations in activities of daily life faced by participants. Importantly, physical inactivity can be considered both a risk factor and an outcome because it may be the result of prolonged illness (Watson et al., 2016). The SHARE participants that were included in our analyses were assessed six to seven times over the course of the study. If one assumes that declining health is associated with increasing physical inactivity, including the repeated measures as a random-effects variable would have attenuated the influence of deteriorating health. In fact, our mixed-effects regression analyses showed that physical inactivity meaningfully predicted various health outcomes, thus indicating that physical inactivity may be at least in part accountable for worse health outcomes. Nevertheless, the SHARE data do not allow for causal inferences on whether physical inactivity was the cause of illness or caused by deteriorating health. Gerontological literature has established that physical activity is a major protective factor in preventing or delaying chronic illness. Therefore, physicians recommend that physical activity is resumed or picked up even in the presence of chronic health conditions (Watson et al., 2016).

Smoking did not appear to meaningfully correlate with any unfavorable health outcome and neither did alcohol consumption. Other studies on the SHARE data came to similar conclusions (e.g., Abuladze, Kunder, Lang, & Vaask, 2017). Importantly, we did not examine the amount of smoking here, because we only included a binary item that assessed whether participants had ever smoked. The group that answered “Yes” also encompassed individuals that had quit smoking in the past. It is well-documented that quitting smoking has a positive impact on many aspects of physical and mental health (Critchley & Capewell, 2003Taylor et al., 2014). Thus, the adverse health effects of smoking might have been obscured by the inclusion of these contrasting groups. Alcohol consumption, on the other hand, was measured by current consumption levels. Our findings align with studies that demonstrate no adverse or suggest even beneficial health effects of moderate alcohol consumption in elderly individuals (Balsa et al., 2008).

Participants with higher BMI's exhibited slightly elevated rates of chronic conditions and reported slightly worse self-perceived health. It is important to note that BMI has long been subject to criticism because it does not account for body fat percentage and body fat distribution which are the main drivers of morbidity and mortality due to obesity (Nuttall, 2015). The BMI's questionable reliability negatively affects its capacity to predict health outcomes which might explain the low correlations we found in the SHARE sample. The fact that BMI showed associations with health despite its methodological issues suggests that the correlations we found in the SHARE sample can be considered to represent a bottom threshold of the true association.

Contrary to our expectations, health behaviors did not moderate the relationship between cognitive ability and health. The only notable exception was physical inactivity, but as discussed above, the direction of effect is ambiguous and unfavorable health effects cannot be causally attributed to physical inactivity here because they arguably exacerbate one another. BMI, smoking, alcohol consumption, and work environment risk did not meaningfully interact with cognitive ability. These results suggest that the intelligence-health association cannot be sufficiently explained by environmental and behavioral risk factors (i.e., at least by those that were assessed in the SHARE interviews). Thus, a different mechanism is required to understand the relationship.

In the literature, education has often been found to explain a substantial proportion of the variance in the intelligence-health association (e.g., Ariansen et al., 2015). In our analyses, controlling for education attenuated the association, indicating a moderating influence. Nevertheless, effect magnitudes remained meaningful, and the directions of effects were unchanged. This suggests that education did exert influence on the relationship between cognitive ability and health outcomes but was not sufficient to explain the effect in its entirety.

The intelligence-health association did not decrease when participants' country of residence was included in exploratory regression analyses. This indicates that the relationship is not meaningfully impacted by regional disparities.

An alternative explanation for this remarkably robust association could be found in a genetic factor that influences health as well as cognitive functioning. The existence of a general fitness factor has been suggested before based on phenotypical findings (Arden, Gottfredson, & Miller, 2009Prokosch et al., 2009), but in recent years more evidence from genome-wide association studies has emerged that directly supports this theory. These studies suggest that a substantial proportion of variance in the intelligence-health association can be explained by genetic variation (Hill et al., 2019). High intelligence and favorable health often coincide because the biological bases of these features are located on the same genes. One of the challenges in this line of research is to deal with the question of causality: do genetic variants affect intelligence which subsequently affects health, or vice versa – or are both intelligence and health affected by the same genetic locations (Deary, Harris, & Hill, 2019)? The results we present here lend some phenotypic support for the latter interpretation. However, more research is necessary to clarify the direction of causality.

4.1. Limitations and future directions

In the current study, our goal was to shed light on the association of cognitive ability and health as well as its underpinnings. The SHARE-dataset represents an invaluable source to investigate these questions goal due to its representativeness, comprehensiveness, and longitudinal nature, but only comprises a limited number of cognitive measures. Cognitive function was assessed using four subtests measuring mathematical reasoning, immediate and delayed word recall, as well as verbal fluency. Despite being considered important components of intelligence in most contemporary established models (such as the CHC-model of intelligence, McGrew, 1997), these four indices can be only considered to be proxies of general cognitive ability. Therefore, we were unable to provide results of more fine-grained domain-specific associations of intelligence with health. In future community surveys, it would be beneficial to include more cognitive ability subtests, especially in terms of highly g-loaded tests such as Raven-typed matrices or figural analogies.

So far, in many longitudinal studies investigating the effects of intelligence on health (e.g., the Lothian Birth Cohort studies, Deary et al., 2007), intelligence was assessed at a young age and subsequently used as a predictor of health outcomes in later life. Here, we are limited by the SHARE study design which does not provide a measure of childhood cognitive ability; the earliest point of assessment was at a participant age of 50 years. Therefore, our analyses potentially carry an inherent bias: reduced cognitive ability at an older age may be the result of poor health, not its cause. To reduce the risk of reverse causation, we controlled for age in correlation analyses. Moreover, if low cognitive ability was the result of certain health conditions, cognitive decline can be expected to progress over the course of the multiple SHARE interview waves. In regression analyses, the cognitive development over the course of the study was held constant which at the very least attenuates this effect.

Death or severe illness may have caused participants to drop out of the SHARE study in later waves. Thus, it is possible that health is overestimated in later waves because the participants exhibiting the worst health are likely to have exited the study. This may have had an attenuating effect on the intelligence-health association in our analyses, thus leading to a conservative effect estimation.

In our analyses, we included risk factors hypothesized to moderate the intelligence-health association. Although the SHARE interviews cover some aspects, other potentially relevant factors are missing. For instance, managing individual health needs is a crucial skill that gains relevance over the human lifespan (Gottfredson, 2004). How well one takes care of one's own health includes how diligently treatment regimens are being observed. Thus far, few studies have tracked patients' adherence to treatment plans in relation to their cognitive ability (e.g., Deary, Corley, et al., 2009Deary, Gale, et al., 2009). These are only some examples of many conceivably influential variables on this association which future researchers may wish to consider.

The intelligence-health association is not merely a matter of psychometric interest, but highly relevant to society at large. In plain words, more intelligent individuals have a higher chance of good health and a long life compared to less intelligent individuals. Advances in healthcare equality seem to have little mitigating effects (Gottfredson, 2004). Consequently, this association has been causing disparities in health outcomes and life expectancies along the lines of the intelligence distribution. Scientific inquiry is a necessary first step in the process of addressing this issue, but large-scale public policy solutions are needed to ensure that all members of society have equal chances of a long and healthy life, regardless of their respective genetic makeup. Unfortunately, public health interventions are often limited to targeting health behaviors, such as smoking or obesity, and ignore the deeper nature of the intelligence-health connection that the evidence is pointing to.

People will put their moral reputations at risk to protect their competence reputations by engaging in unethical behavior that signals (false) competence to others

Clark, C. J., Keighley, D., & Vasiljevic, M. (2022). Being bad to look good: Competence reputational stakes can increase unethical behavior. Evolutionary Behavioral Sciences, Jun 2022.

Abstract: Two studies (total n = 1,245) explored the influence of (a) receiving public versus private performance feedback, (b) competing on a team versus solo, and (c) individual differences in team competition participation on cheating behavior. Participants were given opportunities to cheat in an online trivia competition and self-reported their cheating behavior. Meta-analyses of Studies 1 and 2 revealed that participants who believed their performance feedback would be public cheated more than those who believed their performance feedback would be private, and individuals who regularly participate in team competition cheated more than those who do not. We found no evidence that experimentally manipulating team competition (vs. solo competition) influenced cheating. Our findings suggest that people will put their moral reputations at risk to protect their competence reputations by engaging in unethical behavior that signals (false) competence to others.

Social media use bears no measurable relationship to adolescents' cognitive skills; negative effects of social media use may have been overstated in past research

Associations between social media use and cognitive abilities: Results from a large-scale study of adolescents. Stefan Stieger, Sabine Wunderl. Computers in Human Behavior, June 16 2022, 107358.


• Social media use is not substantially associated with results from tests of cognitive abilities and skills.

• Results from random-forest models suggest low importance of all test results compared to known sex- and age-differences.

• Negative effects of social media use may have been overstated in past research.

Abstract: In adolescence, smartphone use in general and social media use in particular has often been associated with negative effects, such as higher anxiety levels and body dissatisfaction. Other outcomes – such as fundamental cognitive abilities and skills (e.g., intelligence, information processing, spatial perception) – have rarely been the focus of research. Here, we analysed data from a large sample of adolescents (12–16 years; N > 12,000) who performed a series of psychometric tests ranging from intelligence, spatial perception, and information processing, to practical numeracy, and compared their test results with their social media usage (average active and passive time per day, problematic social media use). We additionally applied a random-forest model approach, useful for designs with many predictors and expected small effect sizes. Almost all associations did not outperform known age- and sex-differences on social media use; that is, effect sizes were small-to-tiny and had low importance in the random-forest analyses compared to dominant demographic effects. Negative effects of social media use may have been overstated in past research, at least in samples with adolescents.

Keywords: Cognitive abilitySocial media useAdolescentsIntelligenceRandom-forest model

4. Discussion

The results of the present study can be summarised and discussed as follows. The PSMU scale using a Likert-type response option revealed very good reliability, whereas the reliability was very low for the binary response format (see also Andreassen et al., 2013Wartberg et al., 2017). Therefore, Likert-type scales should be given preference for the PSMU scale, at least when used with adolescent samples.

The correlation between PSMU scores and the average time using social media per day was low (∼0.27; see also Wartberg et al., 2017, who also used adolescent samples and found r = 0.34 for pathological Internet use in general). This supports the assumption that adolescents of that age might miss an objective reference frame of which social media use behaviour is acceptable. As long as the core family (e.g., parents) does not provide negative feedback about adolescents’ possible social media overuse (or even show similar social media behaviour themselves), adolescents will probably not state problems in the PSMU scale although they may use social media for a substantial amount of time on average each day. Additionally, using a parent-form of the PSMU scale or non-parametric measures of social media use (e.g., objectively assessed social media usage behaviour by smartphone apps; e.g., Ellis et al., 2019) might be a good approach for future research.

Past research has found evidence that it makes a difference whether one uses social media actively (e.g., chatting, sharing photos) or passively (e.g., browsing, reposting messages, looking at content; Escobar-Viera et al., 2018Thorisdottir et al., 2019). In the present study, some support for this assumption was found, but differences were of very low effect size (see correlation differences in Table 1, sixth column). Girls were significantly more likely to be actively using social media compared to boys, whereas boys were significantly more likely to be passively using social media compared to girls. Furthermore, active social media users had slightly lower verbal intelligence, whereas passive social media users had slightly higher scores, which is rather counterintuitive. If adolescents’ active use of social media by writing texts and so forth is associated with positive aspects, then we should not expect a negative association with verbal intelligence. Further, although past research found that social media use reduces working memory short-term (e.g., Aharony & Zion, 2019), it does not seem to have long-term effects because the association between active social media use and the long-term memory was, although in the expected direction, of tiny effect size (−0.021; see Table 1) and of marginal importance (third least important predictor in the RF model; see Fig. 1).

Although we cannot draw conclusions about the causal directions of the found effects, in general, the effects themselves were of very weak effect sizes and, compared with each other, the importance of effects mostly did not outperform known demographic differences, such as sex- or age-differences in social media usage (e.g., Coyne et al., 2020). For example, when it comes to the time social media is used actively per day, the association with fluid intelligence (figural) had only about a quarter of importance compared with the sex-difference between boys and girls (girls using social media more than boys). Or put differently, being a boy or a girl is by far more impactful on differences in active social media use than the effect found for figural fluid intelligence.

Furthermore, we found no evidence of any substantial association with adolescents’ intelligence, spatial perception, information processing, technical understanding, creativity, spelling skills, and vocabulary. The only exception was practical numeracy, where we at least found effects similar in effect size to demographics, such as sex-differences (see Fig. 1). Adolescents with higher social media use or higher PSMU scores had lower practical numeracy ability and vice versa. Because of the cross-sectional design, it remains unclear whether adolescents with lower practical numeracy skills more actively search for social media communication or the other (more alarming) way round; that is, more social media usage leads to reduced numeracy abilities, i.e., reduced ability to solve simple text-based mathematical problems (e.g., calculation of areas). Although we also found an association between social media passive use and information processing outperforming demographics, the overall explained variance was very low (1.6%); therefore, we did not interpret this result in detail.

However, the results are interesting from another point-of-view, namely the impact of social media use on cognitions in general, such as the ability to concentrate, hold attention, keep information in memory, and executive functioning (for a review, see Wilmer et al., 2017). Previous research has found evidence that even short-term interaction with smartphones can impact ongoing cognitions by impairing the ability to concentrate or distort attention (Wilmer et al., 2017). For example, one oft-described aspect of smartphone usage in everyday life is multitasking (Judd, 2014), which can have negative effects, such as delayed completion of primary tasks (e.g., Leiva, Böhmer, Gehring, & Krüger, 2012, September) but also positive ones, such as better task-switching abilities (Alzahabi & Becker, 2013) or better multisensory integration (Lui & Wong, 2012). Furthermore, past research has found that even the mere presence of a smartphone can reduce cognitive capacity, resulting in lower scores on intelligence tests (e.g., working memory, fluid intelligence; Ward et al., 2017) or reduced task performance, especially for tasks with high cognitive demands (Thornton et al., 2014).5 Similar studies exist about children doing a school test (Beland & Murphy, 2016Levine et al., 2007).

Looking at the pattern of effects in Table 1, the directions of effects largely correspond to these earlier results. Negative significant associations were predominantly found on tests with high cognitive demand (intelligence test, spatial perception test, technical understanding, practical numeracy, spelling skills [although less spelling errors for highly active social media users]), which were all speeded tests with time restrictions. In contrast, significant positive associations were found on the speeded test with low cognitive demand, namely the information processing test (i.e., more correct answers, fewer errors), which uses simple reaction time tasks. Therefore, the effect pattern could also mean that adolescents do not have lower abilities on the tested concepts (e.g., intelligence, spatial perception), but instead have difficulties with the test procedures themselves because they needed to concentrate and focus their cognitions on a specific task under time constraints. This would also explain why these adolescents performed better in the information perception task. Here, multitasking is beneficial: coordinating information from different senses (seeing, hearing) to perform different hand/foot coordination tasks by pressing buttons. Although this might be a possible reason why we found detrimental effects on low vs. high cognitive demand tests, we do not have direct evidence for that based on the current data, though this would be a fruitful approach for future research.

The present study has also limitations. First, some of the measures (2.2.6 to 2.2.10) were developed and validated in-house at VIC and are not published in any peer-reviewed journal. Nevertheless, all measures were developed over several years under the premisses of being valid, reliable, practical, easy-to-administer, and short. Most of them have a clear face validity (e.g., technical understanding, practical numeracy) and a clear and objective test score calculation (e.g., sum of correct answers). Tests are frequently re-standardised as suggested by the DIN 33430 norm.6 Furthermore, in the present study we assessed the time of social media usage subjectively. Because past research found that subjectively assessed usage time does not necessarily correspond to objectively assessed usage time (e.g., through software-based accurate time assessments; Ellis et al., 2019Sewall et al., 2020), future research should try to replicate the present findings by using an objective measure of social media usage. Because past research found lower associations between objective social media use with, for example, well-being (Sewall et al., 2020), it could be that when focusing on the present results, the uncovered low effect sizes could drop. Another limitation comes from the conceptual distinguishing between active and passive social media use, which has frequently been questioned (for a thoughtful discussion, see Meier & Krause, 2022, March 22), as well as the rather unspecific focus of the measure without explicitly differentiating between the broad range of possible behaviours from texting to watching videos. Because all measures were self-assessed by adolescents, we also cannot rule out a possible shared method-specific variance of the used psychometric tests with social media use. Using objective measures of social media use in future research would resolve that issue.

In conclusion, we did not find any substantial negative associations between social media use and the tested concepts. The effects found did not substantially outperform other known effects, such as sex- or age-differences (except a slightly higher value for practical numeracy on PSMU) if at all. To conclude, although past research found negative effects of social media use in early adolescents (<11 years of age; e.g., Charmaraman, Lynch, Richer, & Grossman, 2021) or children (e.g., 4 and 8 years; Skalická et al., 2019; for a review, see Wiederhold, 2019), cognitive abilities and skills of adolescents between 12 and 16 years of age do not seem to be overly affected by social media use.

How Often and Why Do Guilty and Innocent Suspects Confess, Deny, or Remain Silent in Police Interviews?

How Often and Why Do Guilty and Innocent Suspects Confess, Deny, or Remain Silent in Police Interviews? Lennart May, Yonna Raible, Elsa Gewehr, Johannes Zimmermann & Renate Volbert. Journal of Police and Criminal Psychology, Jun 16 2022.

Abstract: This study examines how often and why suspects who have reported being either guilty or innocent remain silent, confess, or deny accusations in police interview situations. Convicted offenders under current probation or parole in Germany (N = 280) completed a questionnaire about their perceptions of up to six specific police interview situations they had experienced in their lifetime. As predicted, more suspects reported having confessed truthfully (64.3%) compared to falsely (4.1%) at least once in their lifetime; and more suspects reported having remained silent in guilty interview situations (58.4%) compared to innocent interview situations (18.4%). Unexpectedly, approximately an equal number of suspects reported having denied truthfully (39.8%) and falsely (40.2%) at least once in their lifetime. The main reasons reported for these statement types were that evidence seemed to indicate guilt (true confessions), suspects desired to end the uncomfortable interview situation or protect the real perpetrator/another person (false confessions), evidence seemed weak (false denials), suspects felt innocent (true denials), they desired to protect themselves (silence while being interviewed when guilty), and they followed their attorneys’ advice (silence while being interviewed when innocent). Findings are discussed in the context of the police and psychological research and practice.


The present study examined the lifetime prevalences, conditional probabilities, and reasons for suspects’ confessions, denials, and remaining silent in police interviews. We will interpret our findings on the three statement behaviors comprehensively and then discuss their scientific and practical implications.

First, as expected, more suspects reported having confessed at least once in their lifetime in guilty interview situations compared to innocent interview situations. The prevalence of false confessions among our sample was 4.1% and slightly below the range between 5.9 and 24% presented in Table 1 for inmates, offenders and forensic patients. However, the corresponding credibility interval in this study includes this range (95% CI [2.3, 6.9]. The false confessions reported here refer to different types of offenses (theft, fraud, assault, robbery, property damage, drug offenses, sexual offenses). The main reported motives for false confessions were to protect the real perpetrator/another person and a desire to end the uncomfortable interview situation. Ending the interview because of an aversive situation can be assigned to the type of coercive false confessions (e.g., Kassin and Wrightsman 1985), and researchers have already given recommendations on how to decrease the risk of this (e.g., Kassin et al. 2010). In contrast, protecting another person belongs to the type of voluntary false confessions (e.g., Kassin and Wrightsman 1985). Whereas this is a frequently reported reason for false confessions, we know of no literature focusing on how interviewers can detect and minimize voluntary false confessions in order to protect another person. This could be a line for future research.

The lifetime prevalence of true confessions in this study (64%) falls in between the wide range of the four self-report studies examining true confessions (28 to 92%; see Table 1). Also, the true confessions reported here refer to different types of offenses, with most being for theft, assault, and drug offenses. The most frequently mentioned reasons for true confessions were that the evidence seemed to indicate guilt and the suspect’s feeling of guilt. This result is in line with a review by Moston and Engelberg (2011) showing that the strength of evidence is a major predictor for a confession, and the meta-analysis by Houston et al. (2014) who found that true confessions were associated with the suspects’ emotional reactions to the interview and their perceptions of the evidence and their guilt. However, suspects also frequently reported the hope to get a lower sentence as a reason for true confessions. This was also a frequently mentioned reason for false confessions. It indicates that suspects consider the perceived consequences when contemplating confessing (on the effect of consequences on confession decisions, see Madon et al. 2012).

Second, as expected, more suspects reported having remained silent at least once in their lifetime in guilty interview situations (58%) compared to innocent interview situations (18%). To the best of our knowledge, this is the first study to examine the reasons for being silent in guilty and innocent interview situations from the suspect’s perspective. Here, we want to highlight two results: First, the vast majority of suspects reported having remained silent at least once in their lifetime in guilty interview situations because they wanted to protect themselves against misuse of any statement they made. Furthermore, numerous suspects reported that they remained silent at least once in their lifetime in innocent interview situations because they generally do not make statements to the police. This reflects a rather critical picture of the police, and further research on this needs to follow. Second, the attorney’s advice to remain silent was a frequently reported reason for being silent at least once in a lifetime in both guilty and innocent interview situations. Future research could involve attorneys in order to understand the considerations and decisions of suspects and the interview interactions in more detail.

The lifetime prevalence of false denials (40%) was about the same as that for true denials (40%). This result was unexpected and contradicts the findings by Volbert et al. (2019) indicating a higher prevalence of true denials than false denials among forensic patients. Further research should examine this in more detail. However, both studies show that suspects frequently report true denials. Kassin et al. (2003) have argued that a true denial puts innocent suspects at risk: They found that interviewers tried hardest to obtain a confession when they presumed the suspect’s guilt, but the suspect was in fact innocent. From a suspect’s perspective, being innocent and truly denying an accusation can lead to facing an interviewer aiming to coerce a confession. Coercive and accusatorial interviewing, in turn, raises the risk of false confessions (e.g., Meissner et al. 2014). From the police perspective, “truly denying” is a highly challenging statement behavior. The rationale of this is the cognitive mindset of an interviewer: they may launch a suspect interview when they assume that the suspect is guilty. In this mindset, they may assess denials which do not contain conclusive exculpatory information as a sign of the suspect’s guilt. A pitfall here is that they need to distinguish true from false denials, but the ability of interviewers and humans in general to detect deception is poor (e.g., Bond and DePaulo (2006) found an overall accuracy rate of 54%). Probably the only reliable way to assess the validity of denials is by comparing statements with other evidence (e.g., Vredeveldt et al. 2014), but this becomes impossible if corroborating as well as exculpatory evidence is lacking. Suspects most frequently explained false denying in reported guilty interview situations with seemingly unclear evidence, the hope of not being convicted, and the hope of being released from custody. We believe it is fair to assume that these reasons relate to the strength of the evidence. Taking into account the most frequently reported reason for true confessions (evidence seemingly indicating guilt), this indicates the significant role of evidence from the suspects’ perspective.

This study also shows that the suspects made different statements in police interviews, and specific statement behaviors cannot be attributed solely to innocence or to guilt. Taking the reported guilt or innocence as a starting point, we calculated conditional probabilities that allow descriptions of which types of statements the suspects reported most probably for guilty or innocent interviews. Considering the reported guilty interview situations, the probability was highest for remaining silent (40%), followed by true confessions (36%), and false denial (24%). In contrast, for reported innocent interview situations, the probability was highest for true denial (60%), followed by remaining silent (36%), and eventually false confessions (4%). This finding is highly relevant to investigative practice: First, it shows that suspects—when they make a statement—most commonly make true statements (i.e., true confessions and true denials). Second, from a police perspective, the diversity of statement behaviors in innocent and guilty suspects shows the need to conduct suspect interviews in an open-ended manner. Interviewers’ open-ended mindset is at the core of investigating interviewing and is implemented, for example, in the PEACE model (e.g., Bull 2019). The results of the present study provide support for the international effort to introduce and implement investigative interviewing (e.g., European Committee for the Prevention of Torture and Inhuman or Degrading Treatment or Punishment 2019), and generally an open-minded interview approach (Principles on Effective Interviewing for Investigations and Information Gathering 2021).

Finally, for innocent interview situations, the probability that suspects will waive their right to remain silent and deny is higher than that of remaining silent (at least once in their lifetime). In line with experimental findings (Kassin and Norwick 2004), Kassin (2005) has assumed that innocent suspects waive their right to remain silent because they may (a) trust in the fairness of the justice and legal system and expect that their innocence will be believed if they “just tell it like it happened,” and (b) believe that interviewers will be able to read their thoughts and emotions and hence will “see their innocence.” In the present study, the most and exclusive reason for denying in innocent interview situations was the suspects’ explanation “I was innocent.” This underpins Kassin (2005) claim that “innocents put innocence at risk,” because waiving the right to remain silent is an essential antecedent for false confessions. Scherr et al. (2016) also found that suspects’ willingness to waive their rights and deny an offense increased with the strength of their just-world beliefs. However, from a police interviewer’s perspective, the situation is different: Interviewers may conduct suspect interviews when they presume some degree of guilt. Thus, they may assume that the suspect is guilty, assess remaining silent and denying (when no other evidence for cross-checking is available) as an indicator for their guilt, and aim to overcome this and collect confessions. Differently put, remaining silent or denying when being innocent can lead to a risky interview situation with biased perceptions and assessments and coercive interviewing by the police interviewer. This, in turn, can result in false confessions by suspects.


This study is based on retrospective self-reports that have some methodological limitations (e.g., social desirability, cognitive biases, remembering specific events out of multiple similar events, estimated frequencies of events), and we had no information with which to validate the participants’ self-reports (e.g., about their status of being guilty or innocent). These limitations hold true when surveying inmates (e.g., Gudjonsson and Sigurdsson 1994) but also police investigators (e.g., Kassin et al. 2007). Nevertheless, suspects are clearly central to suspect interviews, and their perspectives provide crucial information on them. Second, the current nonrepresentative sample limits the generalizability of the results (e.g., all participants were from one German federal state, German-speaking, without extensive cognitive disabilities). Third, the number of false confessions was small, and this limits the precision of the findings on the reasons for confessing when innocent. Future studies should remedy these limitations by including (a) more and a wider range of participants (e.g., persons from different German federal states, non-German speakers, suspects with cognitive disabilities), and (b) more information about the interview context (e.g., duration, location, persons present) and the personal characteristics of the suspects (e.g., mental health).