How Health Insurers and Brokers Are Marketing Medicare
Our data consist of 1,267 unique ads and 643,852 total airings, representing all English-language TV ads focused on Medicare that aired on broadcast television or national cable between October 1st, 2022 and December 7th, 2022. While the open enrollment period does not begin until October 15th each year, sponsors are permitted to advertise beginning October 1st. Ads were supplied to the Wesleyan Media Project by Vivvix (formerly Kantar) CMAG along with additional information on the timing of the airing, the media market, TV station, and broadcast network where it was aired, and the TV programming shown on the same station during that time slot. The Wesleyan Media Project compiled and processed all data flagged by Vivvix CMAG as relevant to our study, including ads classified as “Medical & Dental Insurance” or “Insurance Agencies & Brokers,” ads sponsored by the Department of Health and Human Services or AARP, and any ad with Medicare in the title.
In the first phase of processing, the Wesleyan Media Project used optical character recognition (OCR) to extract the textual information on screen and automatic speech recognition (ASR) to extract the audio into text for each video. Ads that did not contain the word Medicare in either OCR or ASR were removed from the set. The Wesleyan Media Project then assessed the degree of text similarity from the ASR data to identify advertising that was greater than 90% matching in audio content. Ads with 90% or higher similarity were grouped and only one ad per group was flagged for human coding (described below), along with any ads that did not match with others. Finally, the Wesleyan Media Project identified and removed Spanish-language advertising that ran on Spanish-language networks Telemundo, Unimas, and Univision. A secondary step to filter out Spanish-language advertising was conducted in the human coding stage. This left us with a final analytical set of 1,267 unique ads.
KFF and the Wesleyan Media Project then developed a coding instrument to analyze ad content. Key variables included in the instrument were those relevant to the evolving policy context surrounding Medicare marketing, such as the sponsor, Medicare products, and plans represented in each ad, as well as messaging on plan features, slogans or repeated phrases, activities and characteristics of people shown in the ad, websites and contact information, and use of the Medicare name, logo, or card.
Ten undergraduate student coders received training and completed four sets of practice ads prior to beginning live coding. A random sample of ads (20%) were double coded to assess intercoder reliability, and all Krippendorff’s alpha statistics were calculated for each variable. All results described in this report all met the 0.7 threshold for acceptable levels of reliability, with some exceptions noted below. For ads that included fine print, coders were asked to indicate whether they could read the fine print without pausing. While coder impressions are noted where relevant, this variable is not scientific, as monitor sizes vary, which affects readability. Coders were also asked to flag any ads that stood out to them as surprising, misleading, or notable in some other way and offer their qualitative impressions.
KFF analyzed the frequency of ad characteristics in the data as a whole, as well as by sponsor type, advertised product, and media market. Insurer sponsored ads were also analyzed by firm. KFF further reviewed a sample of ads in their entirety, including ads for which coders had given qualitative comments, to identify relevant examples of common messages, elements, and themes. Messages related to in-home support services, caregiver support, and bathroom safety devices were categorized as “other supplemental benefits” by our coding instrument. These were identified by coders in a free text field and reviewed by KFF as described, but were not formally assessed for intercoder reliability. Likewise, some characteristics, such as the message that beneficiaries were required to pay the standard Part B premium, appeared too rarely to meet the 0.7 threshold for acceptable intercoder reliability. In these cases, we noted that the characteristic was rare, but were not able to report its frequency with precision.
We focused much of our analysis on the substantial majority of ads promoting Medicare Advantage, the most common product advertised during the period.