Data Science To The Rescue: Tackling Real-World Problems With Analytics
Credit: Unsplash
By Professor Eric T. Bradlow (@ebradlow) (GBK Collective and the Wharton School)
Data science has the power to solve real-world problems and create social good. Earlier this week I had the honor of moderating the University of Pennsylvania’s latest Inspiring Impact virtual series. The panel, entitled "Data Science To The Rescue: Tackling Real-World Problems With Analytics” featured Penn faculty experts Dr. Raina Merchant, Prof. Desmond Upton Patton and Prof. Greg Ridgeway, each of whom are harnessing data to solve different problems – from improving health care outcomes to identifying crime patterns to mitigating the risks of social media.
Below, I share some of the key takeaways from our conversation – from the power of predictive analytics to the impact of ChatGBT and generative AI.
Better Data, Better Outcomes
Despite some criticisms, data science has enormous potential beyond business, complicated algorithms, and computer programming. As a Vice Dean of Analytics at Wharton and co-host of the SiriusXM Wharton Moneyball podcast (with my fellow Wharton Professors Shane Jensen, Cade Massey, and Adi Wyner), we get to interview leaders across professional sports every week about how they are applying data and statistics on and off the field. To me, sports is a trojan horse to talk about all kinds of interesting analytical problems that business leaders grapple with every day.
In my role at Penn, I also have an opportunity to review studies that relate to social impact, and have always been fascinated by the use of data science to solve some of our toughest societal problems. At the heart of solving any problem with data science is asking the right questions – unlocking new ways of understanding trends, in order to turn those insights into practical solutions.
To kick off our panel, Prof. Greg Ridgeway, Chair of Criminology and Professor of Statistics and Data Science, Penn Arts & Sciences, discussed how the explosion of data now available has impacted his research on crime and justice. “In addition to traditional methods like surveys, data can now be collected via new processes and sensor networks. Buried within these vast data sets are discoveries waiting to happen about the successes and failures of our systems.”
By analyzing data on drug market transactions, police stops, and use of force, Prof. Ridgeway has been able to identify and address problems in the criminal justice system, helping police departments to reduce use of violence by officers.
Prof. Ridgeway's use of data to predict police use of force is a powerful example of data science's potential impact. By selecting important variables and teasing apart their interactions, he was able to identify risk factors for certain outcomes and more accurately predict them to make better decisions.
“By analyzing data on drug market transactions, police stops, and use of force, Prof. Ridgeway has been able to identify and address problems… helping police departments to reduce use of violence.”
Experimentation and A/B Testing
Another area we spent time discussing on the panel is the importance of ongoing experimentation and random A/B testing. For example, Dr. Raina Merchant, Vice President and Chief Transformation Officer at the University of Pennsylvania Health System, discussed how the Center for Digital Health uses non-traditional data sources like social media to approach healthcare problems in unconventional ways.
By collecting and analyzing data from digital platforms, Dr. Merchant and her team are able to understand individual health statuses and predict changes. At the population level, they study broad trends and messaging distribution to better understand and combat misinformation about vaccines and other health issues.
"For some individuals, having huge amounts of data with very little signal is not as helpful, while others have limited data that is incredibly rich and useful,” shared Dr. Merchant. “There’s no good playbook for how you evaluate or score the quality of the data, so [A/B testing] is something we factor a lot into many of our predictive models.”
Overcoming Bias
As we explored the power of data science in creating social change, the challenge of bias came up repeatedly. We’ve all read studies on the harmful and sometimes discriminating effects of AI and machine learning – from mortgage approval biases to healthcare and public school decisions. The question of how we can use data, algorithms, and AI to improve health equity outcomes while avoiding bias based on gender, race, ethnicity, income, and other factors is more pressing than ever.
Desmond Upton Patton, a professor with joint appointments at the Penn School of Social Policy and Practice and the Annenberg School for Communication, spoke about his work on the relationship between social media and gang violence. By leveraging data science and AI language models, Patton's research helps identify ways to mitigate the risks of social media and its impact on society with the help of social workers, communities, and youth.
However, one of the crucial challenges his lab faces is interpreting social media content without bias. Given the language used online can be difficult to decipher using automated algorithms, Prof. Patton's lab hires young people as domain experts to translate and contextualize posts, providing a fuller understanding of the content.
“We spend an enormous amount of time on the annotation side of data science to extract the meaning of posts, both textually and visually,” said Prof. Patton. “We are training algorithmic systems to find content that might be incendiary.. To counteract bias, we hire young people from the communities in which the social media data comes from as domain experts to help us translate and contextualize these posts.”
The Power of Predictive Analytics
Next, we discussed predictive analytics and how each panelist uses data science to improve impact. I often tell my students: “If you want to just do okay at a job, you build a measurement model. If you want to get promoted, you build a model that allows for prediction and optimization.”
In the business and marketing world, predictive analytics can be applied across the entire customer journey – from awareness to consideration, to purchase – to better understand and predict customer needs in the moment, while also preventing outcomes such as churn they want to avoid.
As our panelists shared, predictive analytics is also transforming their industries. From Dr. Merchant's work in mental health to Prof. Ridgeway's work in criminology to Prof. Patton’s work with social media, the ability to accurately predict and create positive change and prevent negative outcomes is vital.
“Preventive analytics allows us to identify diseases before they happen and improve therapy by providing patients with better data.” - Dr. Raina Merchant
Dr. Merchant's work on predictive analytics in healthcare is an excellent example of the potential for research to directly benefit patients in the clinic. By designing dashboards and gathering patient and clinician feedback on digital data, she and her team were able to accurately capture signals for mental health diagnoses before they were officially recorded.
“Preventive analytics allows us to identify diseases before they happen and improve therapy by providing patients with better data.” shared Dr. Merchant.
The key to success with predictive analytics is to ask the right business questions upfront. Don’t just dive into your data, as you can easily find correlational patterns that you treat as causal, and will lead to spurious findings and poor decisions. And once you take a wrong turn, it can be an expensive and time-intensive ordeal to backtrack. To avoid this, it's crucial to supplement traditional data with new data sources to ensure that the analysis accounts for a broader aperture of what is happening beyond your initial data sets to test your hypothesis.
The Impact of ChatGBT and Generative AI
Next, we delved into the fascinating topic of ChatGPT and generative AI, which has captured the attention of experts across fields. Not only has ChatGBT beat the Arizona bar exam and passed a Wharton Professor’s MBA test – it’s also demonstrated an uncanny ability to imitate human language and thought. I asked each panelist to share what is the state-of-the-art use of ChatGPT within their field and what they see as the biggest opportunities and challenges.
“Not only has ChatGBT beat the Arizona bar exam and passed a Wharton Professor’s MBA test –it’s also demonstrated an uncanny ability to imitate human language and thought. ”
Prof. Ridgeway highlighted the potential of ChatGPT in leveling the playing field for communication including those who speak English as a second language. "I think ChatGPT could be really useful for my students in helping them communicate more effectively. Instead of having to write a novel, they could get to the point up front and adapt to better communicate their policy choices. The second use or objective is for coding data science. It is miraculous how good it is at writing our code. This fall will be the first-semester using ChatGPT and I'm really pushing it to help my students be more productive."
Dr. Merchant highlighted the potential of generative AI to optimize patient care, while also noting the importance of creating the right guardrails. "There are some early data suggesting that generative AI platforms can be more empathetic and even more accurate than physicians in responding to a patient inquiry." However, she also acknowledged the need to carefully consider issues of data privacy and security when using these technologies, stating "transparency to the extent possible would be really important for people having choice and having the ability to understand how their data is being used, where it may be misused, and what the potential risks are."
Prof. Patton stressed the importance of data representativeness when using online or social media data in research. He noted that "we certainly want the most representative datasets possible so that we can be thinking about the best things in this space." While he sees great potential in using ChatGPT for research, he also emphasized the need to carefully consider ethical implications, particularly with regard to issues of privacy and bias, stating "I think we have to question the representativeness of the data."