2019
DOI: 10.3386/w25976
|View full text |Cite
|
Sign up to set email alerts
|

How do Humans Interact with Algorithms? Experimental Evidence from Health Insurance

Abstract: At least one co-author has disclosed a financial relationship of potential relevance for this research. Further information is available online at http://www.nber.org/papers/w25976.ack NBER working papers are circulated for discussion and comment purposes. They have not been peerreviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
19
0

Year Published

2020
2020
2023
2023

Publication Types

Select...
7
1
1

Relationship

0
9

Authors

Journals

citations
Cited by 21 publications
(19 citation statements)
references
References 67 publications
0
19
0
Order By: Relevance
“…45,46 This finding also contrasts with findings from studies of computer-or mailbased information interventions, which showed impacts to be concentrated among younger and healthier populations. 21,22,29 In the absence of intervention, enrollment in Marketplace insurance was particularly low (below 6 percent) among consumers who preferred spoken Spanish and consumers disenrolled from Medicaid. This finding is consistent with prior data suggesting that people with low English proficiency disproportionately experience gaps in insurance and access to care 34,47 and that consumers disenrolled from Medicaid are at high risk of remaining uninsured and losing access to care.…”
Section: Discussionmentioning
confidence: 99%
See 1 more Smart Citation
“…45,46 This finding also contrasts with findings from studies of computer-or mailbased information interventions, which showed impacts to be concentrated among younger and healthier populations. 21,22,29 In the absence of intervention, enrollment in Marketplace insurance was particularly low (below 6 percent) among consumers who preferred spoken Spanish and consumers disenrolled from Medicaid. This finding is consistent with prior data suggesting that people with low English proficiency disproportionately experience gaps in insurance and access to care 34,47 and that consumers disenrolled from Medicaid are at high risk of remaining uninsured and losing access to care.…”
Section: Discussionmentioning
confidence: 99%
“…11,19,20 Several prior interventions sought to improve health insurance decisions via "low-touch" outreach methods, such as presenting information in an automated online choice environment, in an advertisement, or by mail. [21][22][23][24][25][26][27][28] Although these approaches are effective for many consumers, they might not be sufficient to overcome certain barriers to obtaining coverage, such as gaps in health insurance literacy, computer literacy, or internet access. 8,11,12,26,[29][30][31][32][33] Further, consumers in non-English-speaking communities may face language and informational barriers that limit the effectiveness of traditional passive outreach.…”
mentioning
confidence: 99%
“…A leading practical example of this approach is the company Picwell, which combines advanced predictions of the distribution of medical needs with an expected utility model to create personalized scoring for plans. (Bundorf et al, 2019) and (Gruber et al, 2020) show that this approach appears to improve Medicare plan decisions, particularly when the decision aid is provided to insurance agents. Relative to the Picwell-style approach of creating expected utility scores, our approach does not require us to make assumptions about individual risk preferences and displays only the information that theory assumes subjects need to make an informed decision.…”
Section: Introductionmentioning
confidence: 99%
“…Support for the appreciation of automated advice can also be found in an earlier experiment (110): Here, when participants were provided with (correct) human advice and (incorrect) algorithmic advice for a legal case, they very frequently relied on the algorithmic advice, in particular when it was given in production rule form (replicating a previous finding from (111)). Outside of the lab, people were found to be responsive to algorithmic advice when choosing health care plans in a randomized control trial (however, the study does not compare this to responsiveness to human advice) (112). Interestingly, even participants in the control condition (i.e., who did not see the model perform) -of the leading experiment that documented algorithm aversion (84) stated that they were more confident in the model's rather than in human forecasts.…”
Section: Algorithm Appreciation: the Preference For Algorithmic Advicementioning
confidence: 99%