Plan Recommendations in the Medicare Market: A Survey of Medicare Intermediaries

This project surveys licensed Medicare insurance intermediaries (agents/brokers) to better study how intermediaries interact with potential clients and form recommendations. The survey contains an embedded randomized experiment to better understand how intermediaries respond to various client cues and profiles. The survey also elicits agents' beliefs about how much value consumers place on the specific Medicare plan features that distinguish Medicare Advantage and Original Medicare plus Medigap. Finally, we ask about common practices in the Medicare market. The results inform a broader study of how intermediaries shape Medicare plan choice.

2026-06-15

2027-07-01

The primary outcomes are Medicare plan recommendations for different hypothetical client profiles and broker beliefs about how much different consumers would be willing to pay for different Medicare plans and plan attributes.

The Medicare plan recommendation is measured as a binary indicator of whether the intermediary (agent/broker) recommends Medicare Advantage or Original Medicare plus Medicare Supplement (Medigap). We also elicit the intermediary's belief about how much different Medicare beneficiaries (presented as hypothetical profiles) would be willing to pay (in dollars per month) to have Original Medicare plus Medigap instead of a $0-premium Medicare Advantage plan. This allows us to put the value of the plan recommendation in monetary terms.

We use two methods to elicit WTP for Original Medicare + Medigap for a beneficiary profile constructed to be a strong fit for Medigap: (1) a single aggregate plan-level WTP, and (2) an attribute-level elicitation where the beneficiary's perceived WTP is elicited for each MA and Medigap plan attribute and summed. Our primary welfare measure uses the attribute-level build-up (2) since studies have shown that aggregate reports risk omitting plan components the intermediary does not recall when valuing the option. We use the aggregate WTP measure primarily to test how our main randomized statement influences the perceived WTP of the beneficiary. We interpret the attribute-level WTP as the more accurate "true" welfare measure and the aggregate WTP as perceived client demand, however, we will report both measures.

Secondary outcomes include the intermediary's assessment of how well each Medicare plan fits the consumer profile and confidence in their recommendation.

We ask each intermediary to evaluate Medicare plan fit on a 0-100 scale, if they are missing any information needed to make a recommendation, and to report how confident they are in their recommendation on a 0-100 scale.

Licensed Medicare insurance intermediaries (agents and brokers) are recruited through a B2B commercial respondent panel, online survey platforms, online forums commonly used by Medicare intermediaries, publicly available intermediary websites and email addresses, and other online distribution channels. After screening and consent, respondents answer background questions about their business practices and evaluate a series of hypothetical client profiles. We also elicit beliefs about consumers' valuations of different Medicare plan features and common business practices in the Medicare market.

Not available

Randomization is implemented by a computer and is built into the online survey platform.

Randomization is at the individual level.

No

N/A - randomization at the individual respondent level and there is no clustering.

We plan to collect between 150 and 400 observations, depending on recruiting success. We plan to screen for bots, duplicate flags, low-quality or low-effort responses (e.g., always reporting the same answer, always selecting the same right or left-hand side for multiple-choice questions), responses with logical contradictions, and unusually low completion times. These responses will be excluded from the analysis. We also include an attention check question in the second half of the survey. Since this question is late in the survey, we intend to use this as an internal measure of response effort. We believe these responses will still offer valuable information and plan to keep responses that fail the later attention check, but we will report results with and without the failed responses. After screening, we expect the final number of responses to be between 100-300, but will ultimately depend on the response quality and recruitment success. Conditional on funding and recruitment success, we may collect additional responses to make sure we are powered for heterogeneity analysis by baseline characteristics (call center status, tenure) and exploratory analysis by recommendation and knowledge-check performance.

We plan to collect between 150 and 400 observations. After screening (see above), we expect the final number of responses to be between 100 and 300. Treatment and control conditions are assigned evenly (50/50) at the respondent level. We may also conduct some versions of the survey that do not include the randomized experiment so we can add additional questions.

We pre-specify a willingness to pay difference of $100-120/month as our minimum economically meaningful effect. We choose this range because $100 is the smallest plausible Medigap-vs-MA premium difference (but is less than the average minimum Medigap-vs-MA premium difference we observe for our sample); $120 is roughly the average minimum price difference we observe. Given our within-subject design we expect to be powered to detect differences of $100 with a sample size of 100.