top of page

Telling the stories of Michiganders about to lose Medicaid benefits

We used artificial intelligence to assess the human impact
of proposed Medicaid cuts

​The following data showcase provides an overview of a survey conducted by Center for Popular Democracy, Make the Road States, and People’s Action Institute conducted between September 2022 and February 2023. It analyzes the stories of the Michiganders in danger of losing the critical gains in health care coverage made during the pandemic by addressing barriers to enrollment, renewal, and accessing services. The Data & Technology department of Michigan United processed the individual stories using natural language understanding technology (NLU) from IBM Watson NLU. If you've heard of ChatGPT, NLU is the part that understands the sentiment, emotion, semantics, syntactics and context of what you say to it. The other half, the part that creates the response is called natural language generation, or NLG. NLU reveals what's behind the things people tell us in interviews, surveys, polls, social media and more. It is a technology that can change how we interpret first-person responses. See how we used it in some of the datasets to follow. Medicaid improves health outcomes for recipients, improves their financial stability, saves lives, creates thousands of jobs that bolster our local economies, and helps reduce economic and racial disparities in health insurance and healthcare access.  However, while anyone who is eligible for Medicaid is guaranteed coverage, many eligible Michigan residents struggle to enroll in and maintain Medicaid coverage. Even when enrolled, many struggle to get access to the services that they need. This problem is now compounded by the suspension of a pandemic-era rule that made this process a little easier. During the suspension, Medicaid enrollees did not face the usual debilitating barriers to renewing coverage. As a result, the number of enrollees grew from 2.3 million in Spring 2020 to 3 million by 2022. The state's uninsured rate in the state also declined, always a good thing.

Map of Care.jpg

Map of
Our Respondents

On this interactive map, we look at the locations, quality of care, dominant emotion and emotional score for each story.

Tap check the data for a high level look at our at-risk Medicaid users.  

What Did They Need and
Were They Treated Right?

Tap check the data for an interactive, visual guide to respondents' lives, households, needs and opinions.


Demographics, by ZIP Code

In this databook we'll examine details about respondents, their households and more.


Just as we have the right to make a political contribution without fear of retribution, candidates and committees have that right as well.

But considering the history of money, influence and scandal in Michigan politics, we need to compare who gets the money against their record.

Tap check the data for an interactive look at who's accepted the most by aggregate amount and type of contribution.

Tap here for the full dataset


What is
Quality of Care?

The Medicaid program defines quality of care as "the degree to which health services meet or exceed established professional standards and are consistent with current evidence-based clinical practice guidelines." Quality of care also includes the patient's experience of care, such as access to care, communication with healthcare providers, and coordination of care.

Quality of care is human as well.  What Medicaid calls "Access" can be completely different from both kinds of "accessible." "Communication" can be bureaucratic and  difficult to understand.

The following datasets will help you better understand the reasons behind people's quality of care ratings. Not only the numbers but the undiscovered detail about the feelings and experiences underlying their responses.

Quality of Care scatter 2.jpg

The Context of 
Quality of Care

At left are the sentence objects in the stories people told us, arranged as (l to r) negative, neutral and positive sentiment. Vertically, the objects are arranged by the intensity score of the dominant emotion behind the story., with "1" being the most intense.

Tap check the data for an interactive visualization and see whether the quality of care score matches up with the stories and feelings behind it.

Quality of Care by Story1280.jpeg

Quality of Care
by Story

Usually interactive clouds like these use bigger to indicate issues to addent do. In this case, mathematically, bigger is definitely not better.

QOC Keyword Scatter Chart.jpg

A New Look at the Keywords of
Quality of Care 

Tap check the data for a granular but manageable way to understand the sweep of multiple datapoints. 

Medicare Anger.jpg

What Angered Them
about Medicaid

The word cloud at left depicts the angriest of the "Angry" keywords used in stories.


They reflect the object of frustrations--challenges, documents, website, excessive documentation--and other flags that indicate very bad client experiences.

Tap check the data for this interactive feature.

Medicaid Disgust.jpg

What Disgusted Them
about Medicaid

There's anger and sadness. Then there's disgust.


Disgust means trouble.

Tap check the data for an interactive dataset and learn the dangers of people feeling disgusted by your service or organization.

Medicare Keywords QOC and Relevance.jpg

Quality of Care
Keyword Relevance

The honest way to evaluate keywords are by their relevance to the phrase and conversation they're a part of.

The honest way to evaluate Quality of Care is if respondents rate their experience without regard to the possibility of some risk to their benefits.

Tap check the data for an interactive look at where relevance and ratings don't match up.


How They Felt
About Service Timeframes

How long did it take for our respondents to receive benefits and services? How did they feel about it?

Tap check the data for an interactive dataset you can use as a starting point to find out more about people's individual storie

Job Titles.jpg

Who They Liked 
and Disliked

In this dataset we look at entities, namely the job titles of the people they interacted with when applying for benefits or receiving services.

Tap check the data for an interactive cloud.

QOC Concepts.jpg


A concept is a word or phrase that has meanings across different contexts. If someone said "I feel like donuts" they could mean the Dunkin' kind. But if they say it while driving in a snow-covered parking lot, there may be some fun with cars involved.

In this case, the goal is to look for trends and patterns in these higher-level references. Tap check the data for our data.

Categories QoC.jpg

by Relevance

Categories are another classification tool. Tap check the data to see how to classify respondent stories.

bottom of page