Crime statistics measure enforcement, not crime: what else does your data hide?
Chicago police kept showing up at Robert McDaniel's home after an algorithm flagged him as likely to be involved in a shooting, and he later became the victim in two separate shootings. Data strategy consultant Cindy Lin argues that ordinary business data can repeat this kind of harm without anyone intending it. So how does a team build a data culture that drives better decisions and protects people at the margins?
Get a three-layer data culture framework and concrete checks for equity and ethics
Cindy Lin presents her three-layer data culture framework (organization, department and individual) and argues that a strong data culture must also be an ethical one. Using examples from protecting trans students' data, aggregated income figures, crime statistics and a predictive policing algorithm, she shows how data can compound bias and offers practical ways to build equity and ethics into everyday data practice.
“If you think about it, crime statistics are a measure of enforcement, not of crime.”
“And so all organizations use data in some way and have some kind of practices around it, but there's a big difference between having a data culture and having a strong data culture.”
What you will learn
Map your organization's data culture across the organization, department and individual layers of Lin's framework.
Spot how aggregation and limited categories, such as binary gender fields, hide the experiences of marginalized groups.
Question whether a dataset measures a behavior or the process that captured it before using it for decisions or algorithms.
Protect sensitive personal data such as preferred names and pronouns from being shared with people who could cause harm.
Plan a data culture rollout by starting with an assessment of where trust, alignment and skills stand today.
Key takeaways from the speaker
Data Culture is the infrastructure, processes, and practice surrounding how your employees and teams use data to make everyday decisions that run your business.
An Ethical Data Culture is one that ensures that the gathering, protecting, and using of data is done morally and inclusively.
Implementing an Ethical Data Culture helps improve business outcomes.
Details
Level
Intermediate
Format
Talk
Technical depth
Introductory technical
For
Mid-career, People manager, Executive
Length
24:13
Published
June 2023
Questions this talk answers
What is a data culture, and what makes it strong?
A data culture is the infrastructure, processes and practices around how data is used to make decisions. A strong one supports decisions that are made faster and achieve the intended outcomes.
What are the layers of the data culture framework?
The organization layer needs trusted data, alignment, storytelling and governance; the department layer needs relevant tools and a data community; the individual layer needs strong data talent and curiosity.
What makes a data culture ethical?
Lin defines an ethical data culture as one that recognizes and offsets harm experienced by marginalized individuals and organizations in how data is used to make decisions.
How can aggregated data mislead decision makers?
Lin shows that Asian American median income looks higher than the national median, yet some subgroups such as Burmese Americans earn far less, which would change how resources are allocated.
Where should an organization start when implementing a data culture?
Lin starts with an assessment of how the organization compares to each part of the data culture framework, then builds a plan from the gaps found.
About this talk
We rely on data to run all parts of our business: forecast growth, determine resourcing, improving product offerings, measure performance, and so much more. But do our coworkers and / or employees have the skills and tools to use that data, and as important, do they know how to use that data ethically? As women, we have likely experienced the impact of inequitable data systems and so we have a responsibility to model ethical data practices. This talk will cover what an ethical data culture consists of, how we model ethical data practices, and how being an ethical data user can also improve business outcomes.
Protecting trans students' data from unsafe sharing (71 s)
Crime statistics measure enforcement, not crime (64 s)
How a policing algorithm put one man in danger (53 s)
Speaker 1: Awesome. Hello. Give folks just a few seconds here to hop on. I have a full 20 minutes of content, so I'm just going to get started pretty shortly here. Awesome. Great. Well, thank you so much for coming to Modeling an Ethical Data Culture here at the Chief in Tech Summit at the Women in Tech Global Conference. My name is Cindy Lin, pronouns she/they. I am first and foremost the principal at Cindy Lin Consulting. Here at Cindy Lin Consulting, I believe in empowering the impactful ethical and data-driven decision-making that grows organizations.
I offer services to the end in data culture, data product management, and business intelligence, and I'm also available as a fractional data leader. I'm also a member of the speaker team at LA Tech for Good. We work at the intersection of social impact and technology. We foster social change by teaching equitable and ethical data practices so that we can empower individuals to use data more responsibly, building a more equitable world. I first became involved with this organization after attending their workshop on this topic in December of 2020, 2020. Since then, it's really changed how I look at data. One of the things that I'm committed to is having a positive impact in my day-to-day data practice. Let's get started. For those of you who are new to the concept of a data culture, I think all organizations have a data culture because it's just
the infrastructure, processes and practices around how data is used to make decisions. And so all organizations use data in some way and have some kind of practices around it, but there's a big difference between having a data culture and having a strong data culture. And so a strong data culture is one that is used to make strong decisions. Decisions that are made faster, that have the outcomes that they want. And so I'm going to talk to you about how you implement a strong data culture a strong data culture so you can get that outcome using the data culture framework.
So the data culture framework is one that I developed and that you see at a basic level here. It has three different layers, starting with your organization and then looking deeper into the department and at the individual level. And so there must be some work around building out each one of these different layers, and there must be investment in that.
And for each of these different layers, there's also certain principles that I think are really important when it comes to building out this culture, this framework. The first at the organizational level is that you must be building trust, instilling trust in the data that's being presented. And that it's important that data is accurate, that's being shared, so that it can be used to align your organization against what is important towards that everyone's working towards, what is the goals that you're working towards.
So leaders need to make sure that they are good at storytelling so that they can connect what they're seeing with the data, with the reasons behind it, so that your team understands this is what I'm experiencing and that's how it's getting reflected in the data that's being presented. And also leaders need to make sure that there's strong data governance, meaning that how data is accessed and analyzed has been standardized across the company, so there's not three different ways of calculating something like an engaged user.
There's one definition that everyone uses across the board. So that's at the organizational level. On the departmental level, you need to be building tools that are specific and relevant to the departments that you're building them for, so that they are integrated and processes that the team is already employing to run their operations. And then using those tools within that process, you're building a community of team members that are comfortable using data and their day-to-day lives and using that to communicate with each other and having a strong having their own language in which they engage with data and their work.
And finally, at the individual level, you need to be making sure that you're recruiting and developing strong data talent, so that people have the skill set in order to go about analyzing the data that they are looking for. You want to make sure that you're sparking curiosity within those individuals and making sure that they are asking questions that ultimately getting the answers to will help them make better decisions.
And ultimately, if you do all of these things and create a strong framework for data culture within your organization, you're going to empower individuals to be able to answer their questions themselves and ultimately make those strong decisions. So that's the data culture framework. If you're interested in learning more, I've actually written a whole blog series about it.
In progress, they come out on Thursday, so there's much more on these specific topics that you're more than welcome to look at, and that's on my website. You'll see that at the end. And so if you're interested in implementing a data culture, a strong data culture within your organization, I have some ideas of data culture projects that you might consider within kind of structured in those three layers of the framework.
I'm not going to talk about all of them. I'm just going to talk about one each. At the organizational level, you might consider something like what I call the data gem, which is something that I implemented at a previous organization with my team, where we presented a once weekly meeting that was available to everyone in the company to attend on some data topic.
Usually, it was us answering a question that a fellow team member in the organization had asked us, and so not only showing them the answer to that, but showing them how we got there so that they felt like they could easily see how you go from A to Z, right? At the team level, I think operational tools that are powered by data can be especially useful and demonstrate the power of something like a data culture.
In tools like a client health score or machine status. These are tools that help your team understand where they can spend their limited time and have a big impact. If your tools are analyzing all of your clients and saying, these are the ones that are in most need of attention, that's going to be really helpful for your team that is spending all this time trying to figure out what to work on.
So something like that can be really useful. And finally, at the individual level. The one I want to highlight here is data coaching for managers. So this is really helpful for organizations that are in growth mode, maybe promoting from within strong individual contributors to managers. Oftentimes I've seen that those individuals are excited to manage but don't necessarily have those strong data skills to be leading.
And so are failing to make things like ROI or business cases to the executive team. And so don't get promoted further up the ladder because they're not able to kind of make that connection. So that's why I think things like data coaching can be especially useful for those individuals. So these are just some examples of data culture projects that I think can be impactful.
And if you have your own ideas, I'd love to see them in the chat. So, you know, if you go about implementing a data culture, you start to see a strong data culture form at your organization. You'll start to see that data is powerful and some guy once said, or woman, depending on which university you're living in with great power comes great responsibility.
And so it's important for us to think about the kind of impact that we're having when it comes to using data. And so I think that, you know, as we look at the societal issues that surround us day to day, it can be embedded in how data is analyzed and presented and whether intentionally or not, it can compound existing biases and continue to limit access to people already at the margins.
And so I want us to commit to using as a tool to underline how deep oppression goes and create solutions to ensure justice rather than just letting it happen and not paying attention to it. So to this end, I want to cover two topics that are two concepts that I think are really critical to understand as we start to talk about an ethical data culture.
The first is data equity. This definition here really resonates with me. Data ethics refers to the consideration through an equity lens of the ways in which data is collected, analyzed, interpreted, and distributed. It underscores marginalized communities, unequal opportunities to access data, and at times their harm from data's misuse. So obviously it underscores, this quote underscores not just the impact that data equity has, but also the different areas in which you should be thinking about data.
equity. And I think that's reinforcing the second quote here from the feminist feminist manifest now published at MIT equity is both an outcome and a process. So of course we're working towards a more equitable future, but we need to make sure that equity is built into all the steps that we take in order to get to the equitable future.
So that's data equity. The other concept I want to cover is data ethics. And so this quote comes from Harvard Business School's blog. It says that data ethics encompasses the moral obligations of gathering, protecting, and using personally identifiable information and understanding how it affects individuals. Data ethics asks, is this the right thing to do, and can we do better, by Harvard professor Justin Tingley.
So this is, again, emphasizing how important it is for us to be thinking about how we use data, because it has profound ramifications for the people that were whose data we have collected. And so those are two concepts to keep in mind. And so as we redefine what an ethical data culture is, I define it as an ethical data culture recognizes and offsets harm experienced by marginalized individuals and organizations in the infrastructure process and practices around how data is used to make decisions.
Now, you'll notice that second half is still the same as the initial definition I gave you, but in this one, we are specifically focusing on making sure that we're supporting those marginalized individuals and organizations and contributing to justice and not injustice. So we'll talk about from here, I want to talk a little bit about some of the ways in which these concepts manifest and how we use data so you can be more careful about going forward and using data in your organization and your data culture.
So the first one I want to talk about is data security. This is a manifestation of data ethics, of course, we are probably all familiar with this topic, we probably are very invested in making sure that data remains private and secure. But I think there are more ways of looking at this than you might think, or may immediately come to mind.
And so, of course, this is the insurance of safety and security of personal information. And so I want to present a situation in which you may have some trans students at a school that maybe feel supported, feel comfortable being out in that school, but may have transferred transphobic caregivers at home. This is a reality in which I was working with when I was at a previous organization that focuses focus on care for students.
So, in that space, you know, we wanted to make sure that we are creating a welcoming and open and accepting environment for people to be whoever they were. And so we allowed opportunities for things like capturing preferred names, pronouns and gender stored in our student information systems, or the information system we were working with. And on top of that, there are, it was really critical that we thought about how data was shared to individuals that are in these students' lives that may not be supportive of the students themselves.
And so it was very important that we thought about how we protect that data, how we share that data, and ultimately how do we protect children, because there might be certain situations in which that information being shared might lead to an unsafe environment for those children. So it was really critical that we thought about the ways in which that information was being presented, right?
And so while this is not a typical example of how we ensure data security, this is something that is very real, and thinking about how you share data on the platforms that you're working with, or in other interactions, thinking about, like, not just on the product side, about those communications that might go back and forth. But in case, like, you know, caregivers call in, and someone on the support team uses a preferred pronoun, or uses a preferred name or a pronoun that maybe the caregiver is not aware of.
So there are a lot of ways in which you think about how data can be shared that may be inappropriate. So that's the first one. The next thing I want to talk about is underrepresentation in datasets. Here, aggregation and poor data collection practices can hide the experiences of the most marginalized, so we want to think about that.
The example here I have is looking at median income split two ways. So first, we have Asian American median income in 2018 being about $20,000 higher than the national median. But, so that paints one story, but if you were to split out Asian Americans into more specific ethnic groups, you see there's a much different picture for different groups within Asian America.
And so, you know, you see Burmese Americans are making $36,000 a year, significantly lower than the Asian number, and as well as at the national level. So if you have things like where resources are being allocated using one number versus using another or using one set of data versus another, I would probably be making different decisions on how to re on how to allocate those resources.
So, making sure that your aggregation levels are at the right level is really important here. Another piece of that is making sure, you know, if you're collecting data that excludes certain groups, like if you're collecting gender data that only includes male and female, you're excluding people of different gender experiences. Non-binary like myself, gender non-conforming or trans individuals who may identify a certain way.
And if they're not represented in the data, you can't tell the story of them, you can't share what their experiences are. So that's also something to think of as you're collecting data. On the flip side of that, over-representation in data sets can also be a problem. So thinking about how data is captured can over-represent groups and then use as truth, use as truth rather than as a reflection of that process.
And so we see problems with that, especially in things like crime statistics. If you think about it, crime statistics are a measure of enforcement, not of crime. And so it's a reflection of that law enforcement process. And so as a result, we see things like this, right, where this is pulled from the prison policy initiative, where black and Hispanic individuals are much more likely to experience use of force when interacting with the police and much more likely to be arrested once and multiple times compared to other groups.
So this is definitely of a high concern as you think about what is actually being captured here over policing. It's not the fact that these populations are more likely to do crime. Because if I was a policymaker who was an uninformed policymaker, I should say, and I saw this, I might say, oh, we need to send more police to these neighborhoods that are full of black and brown people.
And that might, that would probably lead to a cycle of additional policing, higher crime statistics, and just the cycle getting repeated. And we'll see this in effect and how detrimental it can be on the next slide. Because as we have all seen, AI, machine learning are at the forefront of our industry now. And I don't think they're going away anytime soon.
And so those are often built, those are built upon data sets. And so if those are built upon core data sets and design, algorithms can go and affect people's lives if they go unchecked for discrimination. And so using that example from before, I have a story to tell. This is a story from Matt Stroud. You can find this on The Verge.
It's called Heat Listed. It tells the story of Robert McDaniel, who was flagged by Chicago Police Department's predictive policing algorithm as being likely to be involved in a shooting. And it didn't list whether or not he would be a victim or a shooter. It just said that, you know, given all the data that we have, he's going to be likely involved.
Chicago PD started showing up to his home repeatedly and the community jumped to the conclusion that he was an informant for the police. And that led to him being involved in not just one, but two separate shooting incidents in which he was the victim. And so you can see how things like bad or data like crime statistics can lead to additional violence when used incorrectly.
So just thinking about how we set up our algorithms to be just and non discriminatory. So those are just some examples or these are kind of the main areas in which we can kind of think about how data equity and ethics present in the work that we do. And so thinking about how that connects back to your ethical data culture here again are some ideas on how you might go about making sure that you're practicing an ethical data culture at the also based on the data culture framework.
At the organizational level, you can think about data ethic metrics that can be used to measure impact at the team level. You can look into integrating data equity tools like the data nutrition labels. That's a tool that gets discussed in the LA Tech for Good data equity workshop. So we'll talk about that more in a second. And at the individual level, maybe you just introduce these concepts during data training so that people are familiar with them.
I think that everyone works with data. And so the more people who are familiar with asking these types of questions, understanding how flawed systems can be if they're not thoroughly understood is can be a real big impact. Right. And so when you're thinking about building out your data culture, everyone is involved with that. And so make sure you're pulling from everyone to build one that reflects their values.
I just want to end with highlighting LA Tech for Good leading equitable data practices workshop. Again, this is a super useful tool. It's the workshop I attended that kind of got me started down this journey of data equity and ethics. And you can learn more at that website or you can email [email protected]. They're an awesome organization and you really the workshop is really great because it gives you you can bring the work that you're you're currently working on and you can apply the tools that they give you.
So you can have, you know, hands on direct feedback on what it is that you're working on. And then myself, I am a consultant, like I mentioned at the beginning, so feel free to reach out if any of these topics are something you want to discuss further. I'd be super excited to learn more about what you're working on and how I can help.
And if data culture is something that you're thinking about. Like I mentioned, I have my ongoing blog series on data culture, which you can find at both my website and on LinkedIn. And I, you know, again, specialize in data culture and data product management and in business intelligence, as well as operating as a fractional data leader for organizations that need services like that.
So if you're interested, feel free to reach out. And with that, I'm over 20 minutes, but I would be happy to answer any questions that you may have. Thank you. Thank you, Jacqueline. A question I often get asked, maybe I'll talk about that in terms of implementation, data culture implementation, is what to work on first. One thing that I, I think has been really helpful is using that data culture framework.
And when I'm kind of getting a lay of the land during that. What I like to do in the first stages of implementing a data culture is just getting an assessment done of what's going on with that organization and how that compares to those different parts of the data culture framework. So I can kind of see, okay, there's, you know, low trust, but everyone is using that data to stay aligned on what the goals are.
So at least there's, you know, consistency in terms of what are the goals for the organization. Whether or not they trust it is a different story. And so that is a really helpful way of understanding kind of where to go with your implementation. And you can, you know, build a plan off of that to, to figure out where you want to go to.
So, any other questions? I'll go back to the slide. Like I said, I'd really love to connect if I'm not already connected with you. I actually recognize some of the names here. But if we're not already connected, please definitely shoot me an email or connect with me on LinkedIn. Love to hear what you're working on right now.
And just, you know, open for a coffee, virtual coffee. All right. Fantastic. Well, I'm going to hop off then. I don't see any more questions. But I hope you have a wonderful day. And I hope you're having a lovely conference. Have a good one. Bye.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
How to build a data culture that makes better decisions without harming people
Cindy Lin pairs a three-layer data culture framework with the equity and ethics checks that keep data from compounding bias.
In short
A strong data culture helps an organization make faster decisions that reach the outcomes it wants, but data is powerful enough to deepen existing bias when nobody is watching. Cindy Lin, principal at Cindy Lin Consulting, offers a three-layer framework for building data culture and argues that it must be designed to recognize and offset harm to marginalized people. Her examples, from protecting trans students' records to a predictive policing algorithm, show where that harm hides.
Key takeaways
Data culture is the infrastructure, processes and practice surrounding how employees and teams use data to make the everyday decisions that run the business.
An ethical data culture ensures that the gathering, protecting and using of data is done morally and inclusively.
Implementing an ethical data culture helps improve business outcomes.
Every organization already has a data culture
Lin opens by noting that any organization that uses data to decide anything already has a data culture. The question is whether it is a strong one. "And so all organizations use data in some way and have some kind of practices around it, but there's a big difference between having a data culture and having a strong data culture." (1:52)
Her data culture framework works on three layers. At the organizational level, leaders must build trust in accurate data, use it to align everyone on shared goals, tell stories that connect what people experience with what the numbers show, and enforce strong data governance so that a metric such as an engaged user is calculated only one way. "There's one definition that everyone uses across the board." (3:50) At the departmental level, teams need tools built for their own processes and a community comfortable using data to communicate. At the individual level, organizations recruit and develop data talent and spark curiosity, so people can answer their own questions and make strong decisions.
Projects that bring the framework to life
Lin offers one example project for each layer. At the organizational level, she describes a "data gem", a weekly meeting her team ran for the whole company that answered a colleague's question and showed how the answer was found. At the team level, she recommends operational tools such as client health scores or machine status views. "These are tools that help your team understand where they can spend their limited time and have a big impact." (6:32)
At the individual level, she highlights data coaching for managers. In growing organizations, strong individual contributors are often promoted into management without the data skills to build ROI or business cases for executives, and Lin has seen this hold their careers back.
From a strong data culture to an ethical one
With great power comes great responsibility, Lin reminds the audience, and societal issues can be embedded in how data is analyzed and presented. She introduces two concepts. Data equity looks at how data is collected, analyzed, interpreted and distributed through an equity lens, and, citing a feminist manifesto published at MIT, treats equity as both an outcome and a process. "So of course we're working towards a more equitable future, but we need to make sure that equity is built into all the steps that we take in order to get to the equitable future." (9:22) Data ethics, as quoted from a Harvard Business School blog, covers the moral obligations of gathering, protecting and using personally identifiable information.
From these she builds her own definition: "And so as we redefine what an ethical data culture is, I define it as an ethical data culture recognizes and offsets harm experienced by marginalized individuals and organizations in the infrastructure process and practices around how data is used to make decisions." (10:07)
Where harm hides: security and representation
Lin's first example reframes data security. At a previous organization focused on care for students, preferred names, pronouns and gender were stored for students who felt safe being out at school but might have transphobic caregivers at home. Protecting those children meant thinking about every channel, including a support agent who might use a preferred name on a call with a caregiver.
Next she turns to under-representation. As the speaker cited, Asian American median income in 2018 was about $20,000 above the national median, yet splitting the group by ethnicity shows Burmese Americans earning about $36,000 a year. Decisions on where to allocate resources would change depending on which number is used. "So, making sure that your aggregation levels are at the right level is really important here." (15:06) Collection choices matter too: a gender field with only male and female excludes non-binary, gender non-conforming and trans people. "And if they're not represented in the data, you can't tell the story of them, you can't share what their experiences are." (15:33)
When biased data feeds algorithms
Over-representation can be just as damaging. "If you think about it, crime statistics are a measure of enforcement, not of crime." (15:58) Lin cites figures from the Prison Policy Initiative showing Black and Hispanic people are much more likely to experience police use of force and arrest, and warns that an uninformed policymaker might respond by sending more police, repeating the cycle.
She then tells the story of Robert McDaniel from Matt Stroud's piece "Heat Listed" in The Verge. Chicago Police Department's predictive policing algorithm flagged him as likely to be involved in a shooting; repeated police visits led his community to assume he was an informant, and he became the victim in two separate shootings.
To put ethics into practice, Lin maps ideas back to her framework: data ethics metrics at the organizational level, tools such as data nutrition labels at the team level, and introducing these concepts in data training for individuals. "I think that everyone works with data." (19:48) She credits the LA Tech for Good leading equitable data practices workshop with starting her own journey. In the Q&A she advises beginning any implementation with an assessment against the framework to see where trust, alignment and skills currently stand.
Watch the talk
"Modeling an Ethical Data Culture" by Cindy Lin (24 minutes): https://www.womentech.net/video/modeling-ethical-data-culture
About the speaker
Cindy Lin is a Product Management and Data Strategy Consultant at Cindy Lin Consulting, where they help companies empower employees through strong internal data cultures and retain customers with sticky data products. Their career began in manufacturing and moved into software, with Product Management and Business Intelligence roles at companies including HopSkipDrive, TSIA, Victorious and Avery Dennison.
Questions this talk answers
What is a data culture, and what makes it strong?
What are the layers of the data culture framework?
What makes a data culture ethical?
How can aggregated data mislead decision makers?
Where should an organization start when implementing a data culture?
Drafted with AI from the talk transcript. Editor review pending.
Frameworks and ideas mentioned
Data Culture Framework
Data Nutrition Label
Data governance
Resources mentioned
LA Tech for Good Leading Equitable Data Practices workshop
Cindy Lin's data culture blog series (her website and LinkedIn)
"Heat Listed" by Matt Stroud in The Verge
Data Nutrition Label
Cite this talk
Cindy Lin. "Modeling an Ethical Data Culture." WomenTech Network, 2023. https://www.womentech.net/video/modeling-ethical-data-culture