63 research outputs found
Dataset for 'Reasoning to rationalize or to be accurate'
No full textDataset for: Wischnewski, M. (2022). Reasoning to Rationalize or to be Accurate? Individuals’ Inclination for Systematic Processing and Reasoning-Motivations in Motivated Reasoning.unknow
Code for 'Reasoning to rationalize or to be accurate?'
No full textCode for: Wischnewski, M. (2022). Reasoning to Rationalize or to be Accurate? Individuals’ Inclination for Systematic Processing and Reasoning-Motivations in Motivated Reasoning.unknownunknow
Factors influencing credibility perceptions of AI journalism: Investigating attitudes, trust in news media, and perceived agency
No full textThis bachelor thesis addresses factors influencing the perceived credibility of AI-generated articles in journalism. Specifically, the influence of media trust, agency, the machine heuristic, and the author on perceived (message) credibility will be investigated
Does polarizing news become less polarizing when written by an AI?
No full textIn this study, we examine how readers perceive the credibility of polarizing news purportedly written by a machine. In particular, we study whether a machine attribution can decrease the polarization inflicted by the self-confirmation bias. To that end, we expect that attitude-confirming polarizing news is perceived as less credible when attributed to a machine than when attributed to a human author. We assume this is due to the lower source credibility of machines and less emotional involvement. In a preregistered online experiment, we presented N = 508 participants with a polarizing news article attributed either to a human author or a machine. The article also either confirmed or disconfirmed participants’ attitudes towards the polarizing issue. Our results show that participants did not differentiate between human and machine-attributed news. Moreover, we found no evidence that machine-attributed news affected the self-confirmation bias. However, we found that, while machine authors were perceived equally competent as human authors, they were perceived as less trustworthy. In addition, we found that the machine attribution induced less emotional involvement in terms of experienced enthusiasm but not experienced anger
Preregistration of the data analysis for the validation of the first wave data
No full textDevelopment and validation of a new scale to measure trust in AI-powered systems
Second Assessment Wave Data Collection and Analysis
Three vignettes will be designed, based in part on the pre-trial vignettes but more elaborate. We aim for a sample of N = 600 participants per vignette. We determined the sample size we aim for by using rules of thumb to estimate what sample size is required for the most complex model we plan to fit to the data.
Our goal is to avoid missing values via strict no-skipping rules in the survey implementation. We plan to carry out a complete case analysis.
The first step is to investigate whether we can assume measurement invariance (MI). If we can, then we are going to fit one model for item scores pooled across three vignettes, if we cannot, we are going to fit three separate models, one per vignette.
We are going to assess MI sequentially, starting with configural, then moving from weak over strong to strict. We are not going to investigate partial MI. We are going to use the single items as indicators.
We plan to conduct the following analyses. We are going to perform them in R with the packages psych, lavaan, and semTools. We are going to use the wlsmv estimator for ordered items in lavaan. For any test statistics and fit measures, we are correspondingly going to use robust options as well.
Item-wise description (per vignette):
• Item-wise distributions with bar plots
• Item-specific means and standard deviations
• Compare with description from first-wave assessment
Correlation table (per vignette):
• Compute correlations between all items
• Compare with description from first-wave assessment
Per vignette, for the trust scales, we are going to compute reliability estimates, using Cronbach’s coefficient alpha and hierarchical omega, using 6 factors.
Per vignette, for the additionally administered items measuring situational factors (i.e., riskiness of the situation, affective responses, control agency, perceived anthropomorphism), we plan to compute internal consistency (using Cronbach’s coefficient alpha) per scale/construct. Compare with reliability estimates from first-wave assessment.
After this scale assessment, we plan to compute appropriate sum scores and carry out correlational analyses together with (appropriate) sum scores from developed trustworthiness scales. The correlation matrices are going to be computed per vignette and compared between vignettes.
We are further going to carry out multiple-group CFA analyses (using the package lavaan and semTools) to assess measurement invariance (MI) between vignettes.
For the possible multiple-group CFA models, we consider and are going to compute the following options:
Correlated-Factors CFA compute but assume to discard
2-level CFA compute but assume to discard
Bifactor (S-1) CFA (Eid et al., 2017) most aligned with our test construction
- Orthogonal variance of all specific factors to GT
- Specific factors amongst each other allowed to correlate
Depending on the MI results, we are going to proceed differently. In case we found evidence for MI, we are going to fit a CFA model to the data from all vignettes together and estimate factor loadings. We are going to correlate the corresponding extracted factor scores with simple sum scores to allow for a recommendation regarding scoring rules.
For the context factors, we are going to fit SEMs to assess their relation to the trust scores. Using all manifest context variables and the factor model for trust scales, we are going to use the regsem (Jacobucci et al., 2016) package to select context variables with notable effects on any of the trust factors, using stability selection (Li & Jacobucci, 2022).
Hypothesis for context factors
• We will explore the relationship to the Big 5 exploratively
• Users’ agency/control correlates positively with the system’s transparency.
• Machine’s agency correlates positively with ability.
• Higher machine agency increases the variance of the factor unbiasedness
• The more anthropomorphic a system is perceived, the more individuals trust it.
• Anthropomorphism correlates the most with the factor integrity.
• The more risky the situation, the less people trust the machine.
• The higher individuals’ propensity to trust (PPT), the higher trust.
• PPT correlates more with general trust factor than with specific trust factors.
• The more participants trust, the more positive (and less negative) affect they experience.
References
Eid, M., Geiser, C., Koch, T., & Heene, M. (2017). Anomalous results in G-factor models: Explanations and alternatives. Psychological Methods, 22(3), 541.
Jacobucci, R., Grimm, K. J., & McArdle, J. J. (2016). Regularized structural equation modeling. Structural equation modeling: A Multidisciplinary Journal, 23(4), 555-566.
Li, X., & Jacobucci, R. (2022). Regularized structural equation modeling with stability selection. Psychological Methods, 27(4), 497
Reasoning to Rationalize or to be Accurate? Individuals’ Inclination for Systematic Processing and Reasoning-Motivations in Motivated Reasoning
No full textThe central aim of this registered report is to investigate possible effects of reasoning motivations on the interaction of cognitive style and motivated reasoning. It is suggested that the effects of cognitive style on motivated reasoning are dependent on the reasoning motivation in a way that an accuracy motivation leads to the attenuation of motivated reasoning due to cognitive style, whereas a defense motivation leads to the enhancement of motivated reasoning due to cognitive style. These relationships are proposed to be tested in an online experimental 2 x 3 between-subjects design.unknownothe
Certified = trustworthy? An empirical study of the effects of AI seals of trust on users’ trust perceptions of AI-powered systems
No full textWe are interested in the effects of AI seals of trust on users’ trust perceptions of AI-powered systems. To that end, we explore the direct effects of three different seals on trust
- …
