iiinova — where your digital self keeps evolving

Nick Byrd, Ph.D. @byrdnick.com · 25/09/2026

Concerns about ubiquitous recording are not new. Decades before “smart” devices or traffic cameras were ambiently recording us, we let other digital systems have our sensitive information. We recall 3 not-yet-surmounted issues with deploying this #tech: doi.org/10.1080/1526...

Katherine Saylor and Nick Byrd raise three long-standing issues related to the deployment of ambient recording or ambient intelligence technology:

(1) institutions tend to be averse to rigorous testing and evaluation, settling for mere pre-post or other research designs that can't assess cause-and-effect;

(2) even when proper pre-deployment experiments happen, it may be impossible to operationalize ethical outcomes that do not favor some vulnerable groups over others;

and (3) standard ethical frameworks assume just two parties (scientist-participant, doctor-patient, company-user, gov-citizen) and thus fail to account for the multiplicity of (often not-yet-consenting) parties that have interests in the sensitive digital records we create and store.

Saylor, K. W., & Byrd, N. (2026). Ambient Intelligence Highlights Familiar Obstacles to Pre-Deployment Testing of Healthcare Innovations. The American Journal of Bioethics, 26(2), 32–34. https://doi.org/10.1080/15265161.2025.2608639

001

Reposted by Nick Byrd, Ph.D.

Zoe Purcell 🕵🏻‍♀️ @zoepurcell.bsky.social · 22/09/2026

Why are people so bad at spotting fallacies in climate (mis)information? And what can we do about it? @bencebago.bsky.social and I found a big role of prior beliefs, evidence against identity-protective reasoning, and a neat way to improve climate argumentation. rdcu.be/8X6KVTwk5pTa

024

Nick Byrd, Ph.D. @byrdnick.com · 22/09/2026

The new #AI analyses of our #openData may also affirm some predictions of #BoundedReflectivism (doi.org/10.1111/meta...) and #StrategicReflectivism (doi.org/10.48550/arX... ): Could low #confidence (or "feeling of rightness") trigger reflection? Seems like it could!👇 bsky.app/profile/jere...

Figure 1 depicts the Bounded Reflectivism algorithm (https://doi.org/10.1111/meta.12534), which includes a node for conflict and low "feeling of rightness" (or confidence). The body of the paper unpacks how low confidence can trigger reflective thinking in both humans and machines (or in human-machine collaborations): https://doi.org/10.48550/arXiv.2505.22987

010

Nick Byrd, Ph.D. @byrdnick.com · 22/09/2026

I'm pumped to see others beat me to a re-analysis of our earliest think-aloud datasets. Long live #openScience and #processTracing! We're also developing AI-mediated, human-validated approaches to public decision transcripts, albeit with slightly different research questions. Promising methods! 🧵

131

Nick Byrd, Ph.D. @byrdnick.com · 21/09/2026

Has ego depletion research been resurrected as “cognitive depletion”? “Study 2 [shows that] glucose levels, can modulate deliberative thinking…” doi.org/10.1080/0144... I have a memory of papers arguing that drinking glucose can’t directly change brain glucose. Is that false?

151

Nick Byrd, Ph.D. @byrdnick.com · 19/09/2026

We're sharing a new #MetaAnalysis at #CornellUniversity to the International Society for #MoralPsychology. I posted about our presentation on LinkedIn, and I hope to add comments about other presentations: www.linkedin.com/fee... #ethics #cogSci #xPhi #bioethics #psychology

Reynolds, C. J., Byrd, N., Brown, C. M., & Conway, P. (in preparation). Reflective Thinking Is More Sensitive To Consequences Than Harm-Avoidance: A meta-analytic process dissociation approach. https://osf.io/kj7cv

A stone-clad bridge, with a stone-clad sign showing "Cornel University" at the entrance, and with blooming red flowers in front of the sign.

Houston pond in the Newman Arboretum with rolling hills of grass, flowers, and trees in the foreground and background.

The Robison New York State Herb Garden with raised beds, walking paths, trellises and more.

000

Nick Byrd, Ph.D. @byrdnick.com · 15/09/2026

As long as the German-bound linguistic arc of the universe tends towards greater clarity, I’m in. But if it tends towards Hegelian incomprehensibility, I’m out.

110

Nick Byrd, Ph.D. @byrdnick.com · 15/09/2026

After the #emdashification of #AI-generated text, I started noticing #LLMs hyphenating terms that I would not have hyphenated: cognitive-science graphic-design public-health Do you (humans) usually hyphenate these? If so, why? Why might #languageModels use such hyphenation?

100

Nick Byrd, Ph.D. @byrdnick.com · 14/09/2026

Maybe the final version is different, but the preprint i can access indicated one of the autism measures resulted in a small difference between the groups. osf.io/6bdkv/download

“Scores on the shortened ASQ were significantly higher among philosophy participants (M = 2.68, SD = 0.54) compared to the general population (M = 2.59, SD = 0.54), t(720.921) = 2.182, p = .029, 95% CI [0.008, 0.156], d = 0.152. However, the effect size was small…”

000

Nick Byrd, Ph.D. @byrdnick.com · 14/09/2026

Worthwhile research question! I’d expect this to be found in a cluster of academic fields that have heavy emphasis on analysis that relies heavily on formal systems (such as logic, math, code, etc.), which is probably not all of academic fields (as far as I know).

120

Nick Byrd, Ph.D. @byrdnick.com · 14/09/2026

If you had to guess whether someone’s a philosopher, what’d be more predictive? Their decisions about thought experiments? Or their scores on assessments of neurodivergent traits? In more than 1000 people, the measures of #autistic (and other traits) were more predictive:

doi.org

Thinking differently: neurodivergent traits and responses to thought experiments in philosophers and the general population

Philosophers have long speculated that individual differences in temperament influence philosophical thinking, yet empirical research has rarely explored the role of neurodivergent traits in this d...

5219

Nick Byrd, Ph.D. @byrdnick.com · 12/09/2026

I do not. Sorry. If someone thinks they could recruit a somewhat representative sample of economists, I'd be open to running the study. :)

020

Reposted by Nick Byrd, Ph.D.

Experimental Philosophy @xphilosopher.bsky.social · 12/09/2026

In some existing work, participants are explicitly asked whether their views have some hallmark properties of belief psycnet.apa.org/record/2026-... www.sciencedirect.com/science/arti... Participants sometimes openly acknowledge that they don't – which makes this seem like a workable strategy

psycnet.apa.org

APA PsycNet

012

Nick Byrd, Ph.D. @byrdnick.com · 12/09/2026

I wish Adam Feltz were here to weigh in. In the meantime, there seem to be correlations between personality and economic ideology, controlling for some confounds: doi.org/10.1017/S000... I’d expect to find lots of correlations between (dis)agreement and certain personality traits (e.g., openness).

doi.org

Personality and Political Attitudes: Relationships across Issue Domains and Political Contexts | American Political Science Review | Cambridge Core

Personality and Political Attitudes: Relationships across Issue Domains and Political Contexts - Volume 104 Issue 1

120

Nick Byrd, Ph.D. @byrdnick.com · 12/09/2026

The philosoohical views that’ve proliferated in the past 50 years may be driven more by professional incentive than by introspection or reflection. Many philosophers less-than-publicly admit that what they accept day-to-day contradicts what they professionally endorse. 2/2

120

Nick Byrd, Ph.D. @byrdnick.com · 12/09/2026

I think it depends on the measure. I think a substantial plurality (30-60%?) of philosophers may *say* or even *feel* that they believe the views they publicly argue for, defend, or subscribe to. However, I expect most of them would behave in ways that contradict their professed views. 1/2

120

Nick Byrd, Ph.D. @byrdnick.com · 11/09/2026

The magnitude and meaning of those philosophy-personality correlations is distinct from their statistical “significance”, of course. In that paper, I framed personality predictors as covariates/confounds rather than as primary or theory-derived explanatory factor.

020

Nick Byrd, Ph.D. @byrdnick.com · 11/09/2026

For what it’s worth, I found some big five personality traits *did* correlate with certain philosophical views (above and beyond other demographic and cognitive factors, N > 700): doi.org/10.1007/s131... 🔓 philarchive.org/rec/BYRGMD

Table 5 of “Great Minds Do Not Think Alike” (Byrd 2023) showing multiple regression predicting various philosophical views from cognitive, educational, demographic, and personality factors.

141

Nick Byrd, Ph.D. @byrdnick.com · 11/09/2026

🍿 Glad to see that hypothesis is being investigated! Will be *very* interested in the results!

221

Nick Byrd, Ph.D. @byrdnick.com · 08/09/2026

And, finally, why are footprint symbols on some floors? Do they #nudge people to take the #stairs more often? A powerful experiment didn’t detect that — even when #footprints contained an environmental or #health message! You can follow Dr. Radtke's lab on RG: www.researchgate.net/lab/Health-P...

Do Footprint Nudges Motivate Stair Use at a University?

Dana Braß, Henry Dicken, Paquito Bernard, Mario Wenzel & Theda Radtke

000

Nick Byrd, Ph.D. @byrdnick.com · 08/09/2026

What else predicts #antibiotic use? Iga Palacz-Poborczyk (from above) also presented a poster about data from Poland. Age, education, income, but not gender predicted variance in use of #antibiotics.   Oh, and faulty beliefs or expectations. Iga's also on RG: www.researchgate.net/profile/Iga-...

Behavioral Determinants of Antibiotic Use: Insights from a General Population Survey in Poland

Iga Palacz-Poborczyk, Julia Kuzminska, Aleksandra Luszczynska, Aleksandra J. Borek

You can visit the website for more: https://care-beh.eu/research-team/

100

Nick Byrd, Ph.D. @byrdnick.com · 08/09/2026

If #AI can enhance diagnostic accuracy, why might some physicians not adopt it? Dr. Kritz et al. found what you might expect, such as #health systems that block #LLMs, insufficient training, and being more than 50 years old. 📃 doi.org/10.2196/80274 👤 www.researchgate.net/profile/Marl...

The poster version of the paper below:

Kritz, M., Holawatsch, E., Behrens, D. A., & Hyll, W. (2026). Barriers, Facilitators, and Intention to Use AI for Breast Cancer Diagnosis: Mixed Methods Study Among Austrian Physicians With and Without AI Experience. Journal of Medical Internet Research, 28(1), e80274. https://doi.org/10.2196/80274

121

Nick Byrd, Ph.D. @byrdnick.com · 08/09/2026

If disagreement improves, say, #socialCohesion or #openMindedness, how can we help people engage in it?   An experiment found planning may help.   But self-reported #intellectuallyHumility predicted being *less* likely to change minds? Follow Dr. Scholz on RG: www.researchgate.net/profile/Chri...

Disagreements with known others: powerful opportunities for change and connection

PREDICTOR
Disagreements with known others (friends, family, colleagues, acquaintances)

Planning: Attitude / Behavior Change (Reflect, rethink, and grow)

Intellectual Humility: understanding and stronger relationships

Study 1
- Online Sample: N=581, representative of Dutch population
- Key survey measures for today:
- Frequency, planning, intellectual humility, perceived discourse quality, outcomes

Study 2
- Online longitudinal experiment: T1 N=487, T2 (T1 + 2 weeks) N=461
- Experiment: 3 conditions (control, planning, planning + intellectual humility, 2 timepoints

Planning increased intention and occurrence.

Odds of reporting >=1 DKO at T2 = 2.14x higher
in planning conditions.

100

Nick Byrd, Ph.D. @byrdnick.com · 08/09/2026

How can we help people understand #risk?   Some evidence favors #iconArrays (that visualize outcome frequencies) But what if we added a patient narrative to those #graphics?   A small experiment found #narratives actually *worsened* factual recall! Follow via www.researchgate.net/profile/Made...

Example of icon array and individual narratives.

Examples of individual narratives:

"Side effects are less important, to me... survival is most important."

"In my case, I would accept the risk of long-term side effects.
Because the flipside was — if I hadn't gotten treatment - I may no longer be sitting in this chair"

"In the end, your health comes before everything, right?! think I would take any chance to get better"

The two-arm experimental design:

20 participants
Arm 1: Version with patient narratives (10 participants)
Arm 2: Version without patient narratives (10 participants)

13/20 (former) cancer patients
2/13 breast cancer

Protocol: Semi structured, think-aloud interviews

Measures:
Knowledge test (6/20 scores "Low")
Newest Vital Sign - Dutch

The group who also received narratives had lower factual recall 4.8 vs 6.8 out of 9, ANCOVA B = -2.14, p = 0.002)

100

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

*AND* rapid testing… (Not being able to correct my typos is one of many reasons I often regret threading on this website.) More posts from EHPS in the pipeline!

100

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

One reason #antibiotics are overused is "diagnostic uncertainty". An #rapidTesting can quickly increase certainty. So @isilcoklar.bsky.social et al. have been talking to clinicians and patients to find ways to increase use of point-of-care, C-Reactive Protein #diagnostics More: lnkd.in/p/e5uV4Yg5

Barriers & Facilitators to CRP POCT Use

10 Theoretical Domains Framework (TDF) domains were identified from GP interviews, spanning all three COM-B components - Capability, Opportunity, Motivation.

Barriers
• Diagnostic uncertainty about CRP POCT validity
• Workload and time pressure in consultations
• Cost, calibration and space for the test
• Risk aversion and fear of litigation
• Patient pressure and low health literacy

Facilitators
• Greater diagnostic confidence and safety
• Structured tools: red-flag prompts, scoring
• Real-time CRP availability supports decisions
• Belief prudent prescribing benefits patients
• Peer-delivered education and prescribing feedback

Patient/Carer Perspectives (WP3)

Triangulation of results/How Patient/Carer Interview Findings Link to the Intervention Design

Component 1 - Blended Learning: A test result makes it easier for patients to accept a non-prescribing decision; has been described as "peace of mind"

Component 2 - Implementation & Audit: Pharmacy testing preferred for access and cost; views on who should deliver testing (GPs/ CPs/nurses) shape the delivery model and referral pathway

Component 3 - Patient Resources: Wide variation in AMR awareness; AMR rarely discussed in consultations; patients want accessible, plain-language materials

Cross-cutting - implementation & equity: State funding seen as essential; individual cost models risk excluding patients who can't afford to pay

Task Group Meeting: Finalising the Intervention

• Expert panel review - all three intervention components retained

• GP training: a short online module followed by a face-to-face CPD workshop with role-play, run as a CPD/CME-accredited* session at an existing ICP (Irish College of General Practitioners) conference if possible (not for the feasibility study)

• Workflow tools: the CRP device and consultation prompts add only 1-2 minutes per consultation; monthly feedback can be built as a Ql audit that counts toward GPs' mandatory professional development requirements

• Patient materials: plain-language resources building on existing HSE materials - the best-received component, seen as low-burden and supportive rather than adding to the consultation

* CPD=Continuous Professional Development, CME=Continuing Medical Education

210

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

Ever feel familiar with or concerned about stuff you don’t understand? Most respondents were (rightly) worried overuse of #antibiotics makes them less effective. But many thought #AMR was a #health condition (not a bacterial trait). Follow Dr. Palacz-Poborczyk at www.linkedin.com/in/iga-palac...

Beliefs about AMR (public survey)

Most people were somewhat familiar with and concerned about antibiotic resistance (even moreso for older people).

Curiously, such familiarity and concern did not correlate with antibiotic use!

Understanding of AMR and AMR-related beliefs (qualitative study)

AMR as a personal health condition
- ...AMR was occasionally framed as an instance when antibiotics are incorrectly "matched" to the patient or as a personal health condition that a patient "has" or "suffers from."
- This phrasing is linguistically noteworthy because it depicts
AMR as an acquired dysfunction of the individual, rather than a characteristic of bacteria.

Associations with the term: 'antibiotic resistance'
(public survey - an open-ended question)

Resistance of the body to antibiotics: 44.1%
Antibiotic ineffectiveness: 28.7%
Antibiotic (mis)use: 20.5%
Bacteria's resistance to antibiotics: 12.8%
Infections' resistance to treatment: 11.8%
Don't know/ no associations: 6.5%

In Polish, the words for resistance and immunity are very similar looking/sounding, which may lead to additional confusion:
- Resistance = opomość
- Immunity = odporosć

Understanding of AMR and AMR-related beliefs (qualitative study)
Attributing the risk of AMR to others

Participants projected the risk of AMR onto vulnerable populations, such as older adults and the chronically ill.

They reported a sense of personal invulnerability to AMR, viewing it as an issue that did not concern them.

110

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

Do people think inherited = untreatable? When asked what one's main concern would be if their #genetics predisposed them to a #health condition, 25% said it would be that they “can’t do anything about it”.   Bianca Cucos et al. think personalized risk messaging could help. doi.org/10.1093/eurj...

Personalized Communication Model for Inherited Cardiovascular
Risk: Lessons from Romania

Bianca Cucos, Marius Geanta, Andrie Panayiotou, Angelos Kassianos, Nikos Middleton

This work was conducted within the PERFECTO project - Preventing thE PReventable - Familial HypErCholesterolaemia paediaTric screening for cardiOvascular health

40th Annual Conference of the European Health Psychology Society | 1-4 September 2026 | Pafos, Cyprus

Lifestyle vs. Genetic Factors: Perception and Actionability

There's a partial understanding of the interplay between lifestyle and genetics, with a tendency to underestimate the role of inherited conditions in raising cholesterol levels.

The sense of resignation points to a lack of education around the available options, suggesting the need of more comprehensive public health messages.

Missed Opportunity in Familial Risk Communication

The disconnection between awareness and action highlights an important gap in how genetic risks are communicated within families.

Public health campaigns need to focus not only on informing individuals about their own risk but also on encouraging them to communicate these risks to family members, particularly across generations.

Testing Initiatives and the Role of Direct Information

Most common answer to, "Who recommended you to test your cholesterol levels?": "No one. I went at my own initiative" (> 36%)

Most common answer to, "What type of information about high levels of cholesterol have you searched on the internet?": "The meaning of blood tests, Iipid panel (total cholesterol, good/bed cholesterol, triglycerides)" (>62%)

110

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

We're often not as #healthy as we feel (and vice versa).   And that asymmetry predicts various #health outcomes. But does it predict #healthcare USE? "Feeling healthier than you are" predicted shorter #hospital stays (but not hospitalization). 🔓Calvey & Laurence: www.linkedin.com/feed/update/...

Subjective Health: In general, how healthy do you think you are?
Excellent
Very Good
Good
Fair
Poor

Objective Health
Excellent (10)
Very Good (7.5)
Good (5)
Fair (2.5)
Poor (0)

For every 1-unit increase in health asymmetry (i.e., becoming increasingly health optimistic), a 20% decrease in the cumulative number of nights spent in hospital over the study period was observed.

Health asymmetry did not significantly predict whether individuals were hospitalised or not during the study period.

Discussion

Limitation: Overnight hospitalisations were self-reported, which may introduce endogeneity with subjective health (SH) responses.

Health asymmetry constitutes an additional psychological resource, beyond the individual effects of Subjective Health (SH) and Objective Health (OH) alone.

Psychosocial constructs (like health asymmetry) may be helpful when assessing lengths or recurring hospital stays (and to a lesser extent the risk of hospitalisation).

Takeaway: How older adults rate their health (and the accuracy of these self-assessments) holds significant prognostic value for adverse health outcomes.

100

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

Should #health #psychology researchers embrace #LargeLanguageModels for #qualitativeAnalysis? The EQUAL folks organized a debate at #EHPS2026 to review evidence, arguments, objections, etc. for/against both sides.  👏🤓 (The pro-#AI position lost ground, but sustained the majority.) ehps.net/equal/

Debate

Motion: "Should health psychology researchers embrace
Large Language Models for qualitative analysis?"

Agenda (60 minutes)
• Pre-debate vote
• Opening arguments (5 mins each)
• For: Sumit Mehra (RIVM, Netherlands)
• Against: Thomas Gültzow
• For: Robert West
• Against: Paulina Bondaronek
• Audience questions and comments
• Rebuttals (4 mins per team)
• Post-debate vote

This debate has been organised by the Enhancing QUALitative health psychology with Al (EQUAL) network
• For more details, see https://ehps.net/equal/
• To join, contact Felix ...

Post-debate votes on "Should health psychology researchers embrace Large Language Models for qualitative analysis?"

Yes (for): 52% (17)
No (against): 30% (10)
Abstain: 18% (6)

120

Nick Byrd, Ph.D. @byrdnick.com · 07/09/2026

Should #AI still be benchmarked against humans? Alex Gillespie’s #EHPS2026 keynote shared years of #research making sense of patient complaints with #LLMs. One upshot: models often reveal what humans *miss* — not all human #alignment is good! Follow on gScholar: scholar.google.com/citations?hl...

Healthcare Complaints Analysis Tool (HCAT)
• NHS received 250k complaints/year
• Minimal analysis or learning; merely responding
• HCAT: coding tool for problem & severity
• 13 validations, adaptations and translations

Gillespie et al. (2016). The Healthcare Complaints Analysis Tool: Development and reliability testing of a method for service monitoring and organisational learning. BMJ Quality & Safety, 25(12), 937-946.

van Dael et al. (2022). Do national policies for complaint handling in English hospitals support quality improvement? Lessons from a case study. urnal of the Royal Society of Medicine, 115(10), 390-398.

https://healthcarecomplaintsanalysis.com

Automating HCAT: Using Embeddings?
• 150k patient feedback items (22m words)
• 134 acute NHS trusts, 2013-2019
• Embeddings: similarity to safety-incident phrases ('was misdiagnosed')
• Validated against human wold standard"
• Predicted hospital-level mortality; staff-reported incidents did not

Gillespie et al. (2023). Online patient feedback as a safety valve: An automated language analysis of unnoticed and unresolved safety incidents. Risk Analysis, 43(7), 1463-1477.

A phrase-by-phrase, AI-scored patient complaint excerpt, revealing a various levels of accuracy.

"After driving all day, my farther got home, and ate a normal dinner." 0.33
"Then he became dizzy." 0.44
"We brought him to Colchester Hospital." 0.41
"The treatment was terrible." 0.41
"They were rude." 0.37
"They misdiagnosed him." 0.54
"They left him shivering through the night, almost didn't allow him to eat all weekend!" 0.38
"We kept a bedside vigil to look after him, clean him, and feed him." 0.38
"While the nurses spent hours gossiping around the nurses' station." 0.32

Defensiveness: Symptom Not Cause
• Staff are told to respond to complaints online
• But, they have no authority or resources to address the issues
• Also, they cannot admit to being unable to address the issues
• Their only option: be polite but evasive and noncommittal

Gillespie et al. (2025). The complaint handler's bind: How organisational constraints lead to defensive responses to criticism. PLOS ONE, 20(6). https://doi.org/10.1371/journal.pone.0325185

Gillespie (2026). Provoking interpretation: Using large language models as devil's advocates. Qualitative Psychology. https://doi.org/10.1037/qup0000374

220

Nick Byrd, Ph.D. @byrdnick.com · 04/09/2026

What is the #healthy = #sustainable heuristic? Article: doi.org/10.1111/aphw... 🔓 At #EHPS2026, @katharinaeichin.bsky.social shared an(other?) experiment where people inferred foods' eco-friendliness from its #NutriScore — less so #EcoScore to healthiness. Preprint: doi.org/10.23668/psy... 🔓

H1: Higher Eco-Score --> Higher Health ratings

Label * Credibility: b = 0.46; Cl: -40 - 1.33

H2: Higher Nutri-Score - Higher Sustainability ratings

And this did not depend on Sustainabilty level!

Label * high sus: b = 0.02; Cl: -0.49 - 0.54
Label * low sus: b = -0.27; Cl: -0.78 - 0.24

Discussion
• Nutri-Score effects sustainability perceptions of food
• possibility of unwanted green-washing effects (see e g. de Freitas-Netto et al. 2020)
• Eco-Score only effects healthiness perceptions of foods in individuals that found the Eco-Score more credible - but the effect is smaller
--> unfamiliarity with sustainability labelling in Austria
--> complexity of sustainability concept and possible scepticism (see e.g Rinaldi et al. 2026)
• healthy = sustainable heuristic holds for less, ambiguous' foods (e.g., vegetables, meat) (see also
--> even perceptions of more clearly sustainable/unsustainable foods are influenced by the Nutri-Score
• food-specific healthiness and sustainability knowledge seem to play a minor role in heuristic processing
--> Limitation: stimuli were perceived somewhat differently in the final sample than established in the pre-test

100

Nick Byrd, Ph.D. @byrdnick.com · 04/09/2026

Can #eyeTracking advance #HealthPsychology? Dr. Athina Manoli shared an experiment suggesting that it can! My smartphone has issues that prevented me from capturing key slides, but you can follow Dr. Manoli on ResearchGate to be alerted when this is published: www.researchgate.net/profile/Athi...

Methods

Participants: 32 adults (22 female), aged 18-61 (M = 32.8, SD = 8.11)

Stimuli
• Greyscale, single-item health-threat & neutral images
• Matched for visual complexity, brightness, contrast & size

Manipulations
• Pair duration: 500ms vs. 1500ms
• Pair type: neutral-neutral vs. health-threat-neutral

Self-report measures
• Short Health Anxiety Inventory (SHAI): Health Anxiety & Negative Consequences
• State -Trait Anxiety (STAI): State & Trait anxiety

100

Nick Byrd, Ph.D. @byrdnick.com · 04/09/2026

*Correction: DO the preprints...? (Apologies for other typos I've missed.) Onward!

100

Nick Byrd, Ph.D. @byrdnick.com · 04/09/2026

What if #AI starts the analysis and humans review? Paulina Bondaronek shared 1) Machine Assisted Topic Analysis (MATA): doi.org/10.64898/202... 2) GRACE evaluation framework: doi.org/10.2196/prep... Dr. Bondaronek seemed unimpressed. Does the #preprints formally compare human and machine outputs?

Comparing different NLP models: LDA, BERTopic, LLama 3, Copilot, DeepSeek

• Each model was applied to the same dataset: "What influenced your mood behaviours and well-being?" (N=1,044)
• Two researchers independently analysed the output using MATA
• GRACE (evaluation framework)

Approach: Text --> AI analysis --> Human interpretation

CODE-BASED TOPIC MODELING: LDA, BERTopic

FULLY AUTOMATED "QUALITATIVE ANALYSIS": Three LLMs

Topic Results

LDA ("higher order conceptual analysis requires interpretation")
- Covid as a catalyst for broad physical, occupational, social, mental health and interactions between these factors
- Influences on and impacts on wellbeing, mood, stress and motivation
- Healthy routines and habits to improve physical and mental health issues
- Reflections on covid measures and their impact
- Demands from family, work, health and Covid

BERTopic (narrower descriptions; lacked depth)
- Impacts from house move
- Bag of everything ?!
- Limitations due to muscoskeletal conditions
- Impacts from social connection (or lack of)
- Missing and worrying about family
- Work and/or childcare stress
- Lockdown impacts
- Impacts of dog ownership
- Positives and negatives of vaccinations
- Nothing topic
- Impacts of cancer diagnosis and treatment
- Pregnancy impacts
- Feelings about the future
- Impacts from work-related social connection (or lack of)
- Housing uncertainty due to relationship breakdown

Fully-automated Qualitative Analysis (LLaMa3, Copilot, DeepSeek)
- no details about versions, settings, etc.
- claim: "structured, but vague, needed refinement" — that sounded (to me) like every qualitative analysis I've encountered, making it unclear (to me) how this is worse than human qualitative analysis

100

Nick Byrd, Ph.D. @byrdnick.com · 04/09/2026

How can underutilized tools advance #HealthPsychology? For example, can #AI enhance theory-driven #interview coding? Sumit Mehra et al's attempt found that when #LLM codes deviated from humans' they were usually MEANINGFUL or even ADDITIVE! And most remaining AI codes were identical to humans'!

Benchmark
• Two experienced researchers double coded the interviews, supervised by a third senior researcher
• Codebook based on the Health Belief Model (12 top-level codes + 41 subcodes)
• 1221 fragments were double-coded and discussed
• Approx. 400 hours of work

Prompt engineering:

On the development set, we varied, e.g.
• Chunk size (2000 to complete transcript)
• Number of example chunks (1 to 5)
• Number of coded text fragments per example chunk (0 to 3)
• Prompt style (instruction, role, chain of thought, etc)
• Language (Dutch versus English)

Temperature was always kept at 0

Sidenote on reproducibility:
• The final prompt was run 10 times
• Less than 1% variation in F-value
• So first run was selected for further analysis.

LLM-coded fragments

Number of interview fragments assigned a code by the LLM = 756

Al-coded fragments that were identical to human benchmark: 224 (30%)
Al-coded fragments that differed from the human benchmark: 280 (37%)
Al-coded fragments that were new compared to the human benchmark: 252(33%)

Number of fragments within the test dataset of 10 interviews that were assigned a code by [the LLM?]

Most deviations from human ratings were MEANINGFUL/INFORMATIVE!

And most of the remaining AI ratings were identical to humans'!

Number of LLM-coded fragments = 756
- Al-coded fragments that were identical to human benchmark: 224 (30%)
- Meaningful alternatives to the human benchmark: 220 (29%）
- Meaningful additions to the human benchmark: 168 (22%)

Total identical or meaningful Al-coded fragments = 612 (81%)!

100

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

Then we had a keynote about the history and future directions of #mHealth from from @profjanewalsh.bsky.social. Promising results #remoteMeasurement in remote areas. But not all interventions are panaceas! A "digital clinician" that seemed more knowledgable was also less satisfying or trustworthy?

Evolution of Digital Health

2000-2007: Wearables, Electronic records, Telehealth iPhone

2008-2013: Consumer activity trackers, Smartphone health apps

2014-2018: Big data & Al, Prescription Digital therapeutics, Social Robots

2019-2021 (COVID): Autonomous Al diagnostics, Chatbots, Remote patient monitoring, Telehealth scale

2022-2023: Al clinical tools, Virtual hospital, Augmented Reality, Digital Humans

2024-2026: Gen Al, Biosensors, Digital human

Virtual Ward in action - COPD
• Oxygen saturation probe
• BP & HR monitor
• Thermometer
• Samsung Galaxy Tablet, 5G enabled
• COPD booklet
• One to one instruction
• Digital dashboard monitored by staff
• Measurements uploaded via tablet to dashboard
• Results (RED-AMBER-GREEN)
• Instant messaging/video

Preliminary Results - COPD
• Assessed >500 patients
• Average length of stay: 6.77
• Patient bed days saved: 500

"Digital Clinician" patients:

• Better knowledge (p<.001)
• Less satisfied with consultation (p<.001)
• Less trust in their education provider (p<.001)

Coleman et al. (2025) JMIR Diabetes

100

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

What about #measurement problems? A handout, presentation, debate, and discussion gave us lots to think about re: #Wicked Problems with #psychology scales Thx T. Cornelius, @matherion.mastodon.nl.ap.brid.gy, @mayabraun.bsky.social, @paulinaganucheau.bsky.social, schenk.bsky.social, and M. Johnston

Measurement issues:

1. Construct complexity: Psychological constructs are often complex and multidimensional, but measures may reduce them to a single score or limited set of items.

2. Poor measurement transparency: Researchers often do not fully report how measures are developed, adapted, administered, or scored.

3. Researcher flexibility: Choices about items, scoring and subscales can influence results.

4. Questionable validity: Established and "popular" scales are not necessarily valid for every construct, population, or context.

5. Inconsistent relationship: Different scales attempt to measure the "same" construct differently.

6. Weak validation: Statistical sophistication cannot compensate for a measure that does not adequately capture the intended construct.

Terminology or framing effects?

This slide shows two plots about a statement about one's "ability to move regularly". If I understand correctly, one item asks about "how CONFIDENT..." and another asks "to what EXTENT...". And each item had rather different relationships with other scale items.

Why do we use scales with equivalent items?

We test for internal reliability
• assumes all items measure the construct equally
• Often with high reliabilities suggesting redundancy

What about other formats of items and scaling? Eg.
• selecting single beat fitting item (Thurstone scaling) or
• ordered structure
--> Eg. Bandura's method of measuring safefficacy

(I didn't fully understand this slide, and maybe not just because I have not seen Wicked. I welcome clarification.)

Do we want to continue measurement in the Glinda way?

• 'In the universe of Wicked, Glinda is not truly evil, ... prioritizing her social status and public approval over doing what is right.....remains trapped in a hollow position of power, universally adored yet entirely isolated after losing her best friend? –Reddit

...there never was a Wicked Witch of the North (in official Oz lore)

Or do we need to think again about the constructs, we measure and the items and scales we use to measure them

100

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

Poolside coffee + fruit breaks are the way! Thanks for choosing this venue for #EHPS2026, @ehpsconference.bsky.social!

A man who vaguely resembles Neil Patrick Harris sitting poolside with a plate of fruit near Paphos Cyprus.

111

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

Maybe it's unsurprising that #trust in institutions like the #CDC is decreasing in the #USA, but what about other countries? Well @gerrymolloy.bsky.social and colleagues found a similar trend in #Ireland (Figure 1 via Kevin below) and #Canada. #publicHealth #Science #medSky #SciComm #EHPS2026 #edu

Two plots exhibiting nearly maximal variance in trust metrics by county.

Left: Country-by-country response distributions on a scale from "You can't be too careful" (0) to "Most people can be trusted" (10).

Right: Country-by-country response distributions on a scale of trust in policians from "No trust at all" (0) to "Complete trust" (10).

Figure 2 Canada: Percent of respondents who affirmed "I trust to a great extent" from most to least trusted:

1. Doctors or healthcare providers

2. Scientists

3. Local/public health authorites

4. The World Health Organization (WHO)

The misinformation and disinformation war on science, scientists and experts is not new:
• "We have to honestly and aggressively attack the universities in this country."
• "The Professors are the enemy"

– US Vice President Vance (quoting Richard Nixon)

Contemporary misinformation and disinformation about health and medicine: Will GenAl save us?

Gerry: I have read on the Centers for Disease Control website that the statement that "vaccines do not cause autism is not an evidence based claim because studies have not ruled out the possibility that infant vaccines cause autism. Should I get my children vaccinated against MMR?

chatGPT 5.2: Yes — based on the best available scientific evidence, you should vaccinate your children with MIMR unless a qualified clinician identifies a specific medical contraindication. First, it's important to clarify something: the statement you're queting is not an accurate reflection of the position of the Centers for Disease Control and Prevention. The CDC's published position is that extensive research shows vaccines do not cause autism. If you encountered wording suggesting otherwise, it may have been taken out of context or misinterpreted.

121

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

The US CDC changed its website's claims about #Autism and #Vaccines on November 19 2025. A high-powered experiment found the change caused differences in the perceived #risk and uncertainty about vaccination, trust in CDC, and reliance on #logicalFallacies (but not vaccination intention). 👇 Thread

Change of CDC communication about vaccines and autism:

1. CONSENSUS-based CDC statement before November 19, 2025
• Studies have shown that there is no link between receiving vaccines and developing autism spectrum disorder (ASD).
• No links have been found between any vaccine ingredients and ASD.

2. UNCERTAINTY-based CDC statement since November 19, 2025

• The claim "vaccines do not cause autism" is not on evidence-based claim because studies have not ruled out the possibility that infant vaccines cause autism.
• Studies supporting a link have been ignored by health authorities.
• HHS has launched a comprehensive assessment of the causes of autism, including investigations on plausible biologic mechanisms and potential causal links.

Method: Experimental online survey in Dec 2025/Jan 2026 with N= 2,989 U.S. participants (quota-representative with respect to age, gender, and political orientation)

Experimental conditions:
--> Control: No CDC statement before outcome measures
--> Consensus-based statement: Screenshot of CDC website used before Nov 19, 2025
--> Uncertainty-based statement: Screenshot of CDC website used since Nov 19, 2025

Outcome measures:
• Vaccination-related effects
--> Perceived vaccine side effects
--> Uncertainty about vaccine safety
--> Vaccination itention (own & child)

Societal effects
--> Trust in CDC, government, scientists
--> Societal polarization: social avoidance & discrimination between pro- and anti-vax respondents
--> Endorsement of science denial strategies (e.g., impossible evidentiary standards, conspiracy-based reasoning)

The two messages differed from control in opposite directions) for the following outcome variables:
- Perceived risk of vaccine side effects
- Perceived uncertainty about vaccine safety
- Trust in CDC
- Reliance on logical fallacies
- Conspiracy-based reasoning

Additional results
• Intention to vaccinate their children was descriptively lower in both CDC statement conditions, but the reduction was statistically significant only after exposure to the uncertainty-based statement
• No substantial effects on societal polarization (social avoidance & discrimination)
• No substantial partisan effects in response to CDC statements (despite significant differences between Democrats and Republicans on outcome measures)
• Despite substational reduction of trust in CDC after exposure to the uncertainty-based statement, trust in government and scientists was largely unaffected

100

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

How can we help people take opportunities for #exercise "snacks"? Enter the "Active Waiting" app: activewaiting.at Now, what if the app gave users #AI tailored recommendations based on a photo of their environment? Faith Young reported that this helped in certain cases — and experts preferred it!

Methods
• In a within-subject field study participants (N= 25) received 600 exercise recommendations across four real-world environments.
• Each exercise recommendation rated on a set of Likert scales for perceived appropriateness, safety, comfort, spatial fit and likelihood of future execution.
• RAPA questionnaire, GATAl questionnaire and post-study interview.
A subset of exercises underwent independent blinded expert evaluation (N= 10).

110

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

David Haag et al found #AI generated behavior change techniques were rated better than those from #healthcare professionals (doi.org/10.1145/3706...). So will such #LLMs perform even better if they generate THEORY-concordant recommendations? A small experiment didn't detect an advantage of theory.

Acceptance is high-but behavior change technique (BCT) use remains unclear

1. LLM-generated JITAls can be highly acceptable: The Last JITAl found higher acceptability than layperson-or clinician-authored suggestions (Haag et al.,
2025).

2. Theory-guided LLMs can improve health outcomes: HealthGuru's BCT-informed dialogue improved sleep duration, activity scores and motivation versus a baseline chatbot (n=16; Wang et al., 2025).

3. Appropriate BCT application remains untested: Outcomes do not show whether the right technique was selected, faithfully realised or tailored to the moment.

Can LLMs select theory-concordant BCTs for the right person and context?

A comparison between two model configurations across 71
JITAl decisions (24 participants × 3 recalled contexts x 2 model configurations)

PERSON-LEVEL INPUTS
• Demographics and socioeconomic
• PA level (MVPA, IPAQ)
• habit strength
• Attitudes towards exercise
• Exercise-related Motivation, intention, self-efficacy, social, norms, habit strength, planning, goal setting, action crontrol
• Personal PA goals

CONTEXT-LEVEL INPUTS
• Time, weather calendar entries, background, location, current activity, phone usage
• Mood (valence, arousal, calmness),
stress
• Momentary PA motivation, barriers for
• Previous PA on the day

No differences in Content appropriateness or Expected effectiveness (between base model and BCT + theory).

But the BCT + Theory did seem to perform better (51% compared to 35%)

100

Nick Byrd, Ph.D. @byrdnick.com · 03/09/2026

And @robertboehm.bsky.social closed with reflections on - the multi-factorial, multi-party nature of health problems like #antibiotic resistance. - a growing need for our #research to refine policies, practices, and norms. Interested? Join @abc-network.bsky.social: www.a-bc.network/involved.html

Robert Böhm describes how many factors, people, and the relationships there-between are involved in antibiotic use. Dr. Böhm also identified organizations who care not just about medical science, but also about our research in decision science, health psychology, etc. For example, Robert is a behavioral scientist (with a Ph.D. rather than an M.D.), but is also the director of the WHO Collaborating Centre on Social and Behavioural Research in Antimicrobial Resistance: https://whocc-sabrar.univie.ac.at/

100

Nick Byrd, Ph.D. @byrdnick.com · 02/09/2026

Then @miroslavsirota.bsky.social shared cross-cultural "terminology effects" on correct answering to questions about #antibiotics. "Infection" was more potent than "Resistance". "drug-" or "antibiotic-resistant INFECTIONS" often caused more benefit than "anti-microbial/biotic RESISTANCE" #MedEd

Experiment 2: preregistered six-country experiment

Between-participants experiment.
- National age and sex quotas.
- Analytical sample n = 3438.

Four terminology conditions:
- Antimicrobial resistance
- Antibiotic-resistant infections
- Antibiotic resistance
- Drug-resistant infections

Countries: USA, Brazil, Nigeria, Turkey, India, Australia

Survey languages: English, Brasilian, Portuguese, and Turkish

All four terminology effects replicated across countries, languages, and education.

The terminology effects did vary by secondary outcome, with "spread" showing very smaller effects for "anti-microbio/biotic resistance" than for "drug-resistant" or "antibiotic resistant" "infections".

Implications for communication and policy

Scalable intervention: A simple wording change improved understanding without any extra explanation.

Policy relevant: Clearer wording increased the perceived priority assigned to government action to tackle the issue.

Suggested approach: Use infection-based terms directly, or pair them with established terminology as explanatory language.

Limitations and suggested research: Real-world materials, behavioural outcomes, and across more languages and cultures

Context: Wellcome Trust 2019; Krockow et al., Commun Med 2023; Grailey et al., Commun Med 2025

110

Nick Byrd, Ph.D. @byrdnick.com · 02/09/2026

What else drives overprescription of #antibiotics? @kevinroche.bsky.social asked Irish GPs. As in many nations, the doctors blamed time pressure, patient expectations, and diagnostic uncertainty. But they also proposed a few solutions! You can follow Kevin at www.researchgate.net/profile/Kevi...

Time pressure: Consultation length and pressure for appointments
"in some ways even though it may not be the correct thing to do it may be time saving to actually prescribe an antibiotic for the cough rather than go through the whole process of kind of explaining the reasons why and the reasons not to"

Patient factors: Patient demand and patient expectation
"you know they come in and say! have a chest infection I'm here for an antibiotic and that's how you start the consultation"

Uncertainty: Difficulty of differentiating between bacterial and viral illnesses in General Practice setongs
"being unsure as to whether or not it's a Viral iliness or a bacterial illness and that because there's no reliable test to say this is 100% viral, you probably over prescribe antibiotics"

Irish General Practinioners's ideas about problems and solutions regarding antibiotic overprescribing:

Clinical knowledge is not the issue: "they have enough clinical knowledge to realise when an antibiotic isn't required"

Communication may be an issue: "...might not be confident enough in explaining to patients why they're not going to prescribe an antibiotic"

Consultation skills: "do a thorough examination because you know you have to get the high moral ground"

Side effects of antibiotics: "and certainly you'd be very liable for if they did for example snap their Achilles tendon courtesy of your ciproxin antibiotic which wasn't appropriate"

Pharmacy: "maybe get a community Pharmacist to come our and explain sort of their side of things, what they've seen happening in General Practice"

Microbiology: "the local microbiologists I think they could also be involved in sort of a GP training"

Consultation skills to ensure
• Good history taking
• Examination skills

''a cough that starts with a sore throat and develops nasal symptoms is highly likely to be viral in origin compared to a productive cough that starts without a sore throat or nasal symptoms"

2. Communication skills to mitigate
• Patient expectation
• Patient demand
• Preserving relationship

"I mean you know 99.99% of the time the illness is going to get better itself and the trick is just not to intervene but reassure people"

Conclusions
• Determinants of antibiotic prescribing in Irish General Practice similar to those found internationally
• Qualitative findings allowed us to identify the challenges for GP trainees in becoming independent prescribers
• Findings can be used to inform curriculum development by GP educators
• Optimising the antibiotic prescribing behaviours of GP trainees before they coalesce can contribute to the ongoing efforts to improve antimicrobial stewardship

110

Nick Byrd, Ph.D. @byrdnick.com · 02/09/2026

Back to our session on #Health #Communication and #AntibioticStewardship: First, @schneali.bsky.social shared results from a new #openAccess paper. 🫣 When providers could *choose* to ignore diagnostically relevant info, many did...and also prescribed more #antibiotics? WILLFUL ignorance? 👇Thread

120

Nick Byrd, Ph.D. @byrdnick.com · 02/09/2026

What about colorectal #cancer screening? Dr. Marcella Bianchi shared results from #Campania, which has outstandingly low #adherence. At least three psychological profiles emerged from non-adherers (photo 2). Follow Dr. Bianchi and the broader #MIRIADE project: www.researchgate.net/publication/...

Where this talk fits into the broader MIRIADE project: "Psychosocial profiling: segmenting non-adherers on the determinants identified" in prior "Qualitative exploration", "Determinant studies", and "Measure development".

Three psychological profiles emerged.

1. Psychosocially resourceful: Above average on every construct; highest on planning, norms, regret and identity. And still not attending regularly.

2. Interested-but-unprepared: Best attitude of any group and good perceived control — but the lowest action planning anywhere in the sample. A motivational-volitional gap, not resistance.

3. Vulnerable and disengaged: Below average throughout, with the most negative attitude of any variable in any cluster. Broad disengagement from prevention.

100

Nick Byrd, Ph.D. @byrdnick.com · 02/09/2026

What predicts being more or less intent to seek lung #cancer screening? Andriana Theodoropoulou (below) shared insights about eligible adults with #EHPS2026 from a new paper: doi.org/10.1093/abm/... 🔓 💬 In addition to standard #surveyMethods, they coded thought lists to enable multiple regression.

Cognitive and affective measures (page 4)

Sample demographics (N - 1190), with descriptions of regression and thought listings analyses.

Table 2 shows logistic regression of screening intention on three sets variables (a la hierarchical regression). The strongest predictors were moral norms (favoring screening) and perceived control (in screening).

Table 4 shows multiple linear regression of (coded) thought types of the cognitive and and psychological variables. A stand-out predictor (above and beyond other factors) seems to be counter-arguments (objecting to screening or its utility).

120