Jonathan Care, Ph.D.

The rules of thumb in null hypothesis significance testing really aren't doing business partners any favors. When we say "we are powered for a 3pp effect, with alpha = .05", an analyst would not be surprised if at scale we saw a 1.5pp effect (a 50% reduction from the BP's expectations...) That's because the minimum detectable effect just means the smallest average treatment effect that would make it extremely surprising if the actual effect was zero. Communicating uncertainty is a real challenge in a world of strict cut-offs, but it is also critical if we want BPs to really understand the actual questions they are asking with a test. (This is one reason why I think that whereas Bayesian approaches are more work on the front-end (defining priors, computationally intensive), the back-end is much easier for a BP to understand, because we can say things like, "Given what we believe, there's an 80% chance that the effect is 3pp or greater.", but I digress...)

2

1 Comment

Sounds like 'we need more government money for shoddy statistics' to me: "The report cited three critical weaknesses affecting statistical agencies, including overdependence on political agencies for financial support, scarcity of manpower and budget, and insufficient decision-making power free from political guidance and interference.   So there’s a shortage of bar graphs and decimal points. Who cares? Researchers answer: Everyone. Unreliable data, especially when presented as reliable, puts the nation’s economic, social, and political future at risk. “Lousy data beget lousy decisions,” said Erica Groshen, senior economics advisor at Cornell University. “It is no exaggeration to say that Americans’ well-being and the vitality of the U.S. economy rely … on the quality of information provided by our federal statistical system.”" [wink wink: it is an exaggeration] #Fearmongering #BureauOfLiesAndSophistry #SeasonalAdjustments #HedonicAdjustments

1

2 Comments

🔢 Currently reading and thinking about how much data's consumed in tabular form -- easy to confuse & difficult to tell a story 🔢 Bureau of the Census Manual of Tabular Presentation (1949) An outline of theory and practice in the presentation of statistical data in tables for publication https://lnkd.in/eD3wt7U9

15

2 Comments

my recent article

1

1 Comment

How can you make sure your benchmarking results are statistically sound? Boriharn K proposes two powerful options: the independent samples t-test and Welch's t-test.

3

Open-Source Software Question 🤔 Imagine a (we'll say hypothetical) organization has a policy that limits the number of acceptable versions of any software so, as an organization, they need to coordinate updates to things like R and Python. Does anyone have experience with this in your organization...with R and/or Python specifically? Could anyone share a policy or point me to a best practice? I don't have much production experience, mostly R&D, and am curious how this situation is typically handled (if it is encountered at all). I'm hoping to find something that aids security and infrastructure folks but doesn't require data scientists, economists, and statisticians to become full-time software developers updating their code every few months to maintain compatibility. Would love to hear people's thoughts on this.

14

21 Comments

"We have too many reports, we need a report to help us summarize our reports." Analyst creates a report with the key metrics. "No - I need extra metrics in the summary report, otherwise there's no context." Analyst adds requested metrics "No - this is too complicated. I can't make sense of everything with all these numbers! I need a new report to summarize the summary report!" And the cycle starts again. As analysts and data scientists, we're told to be partners, not order takers, but at the end of the day our success (aka career progression) is based on how our stakeholders feel about us, and if we go around telling them they're wrong to want ANOTHER summary report with the same metrics, we're not going to do ourselves any favors when the "who should we promote" question rolls around. It's absolutely our job to educate our stakeholders, but sometimes the incentives are set up all wrong for us to have the difficult conversation necessary to do so. Just an observation.

12

1 Comment

LangSmith Evaluations: RAG Evaluation (Answer Correctness) Evaluations can accelerate LLM app development, but it can be challenging to get started. We've kicked off a new video series focused on evaluations in LangSmith. https://lnkd.in/gw9b7nNU https://lnkd.in/gTyvUtcz

4

#datafam have you read The Benefits of Data Visualization for Data Scientists? This tutorial from Megan Spengler at Playfair Data is one that personally resonates with me for many reasons. Early in my career I saw so much hard work get shot down at the pitch because it was poorly presented to stakeholders. As data people it's our job to ensure we present our work in a way that resonates with our stakeholders and makes sense to the audience we are communicating it to. Time and time again data visualization can help get us there. Check out the tutorial from the link in the comments below.

27

4 Comments

It’s #DataViz #SuperBowl season. If you see me posting images of Rick Klein or Steve Kornecki- know I’m not pushing electoral discussion- but I’m #nerding out on mainstream #news data visuals on dashboards. This is a side of me I have to surpress on social media because it gets political REAL FAST in the comments- but I just want to talk about the COMMENTARY and Live stats PRESENTATION y’all! And the maps!!! #datacommunicator True story, when drafting and pitching my #datacommunicator role profile / job description research- I had to make an extremely advanced version as well as a very basic draft of the position before our internal review folks understood what I was talking about. The extremely advanced version was for public speakers on the news like Steve Kornecki, who has gone on to do live data viz presentations at other events like the Kentucky Derby. It’s why I remain patient to see what will come from #VR and #AR layering, as promise has been shown to get to a future Hans Rosling would “squee!” about as #tech and #datascience evolves. #nerdalert Anyone else this data viz nerdy?

5

I am pleased, proud, and humbled to announce the publication of my forthcoming book later this summer Why Data Science Projects Fail, with my co-author Evan Shellshear, PhD. Below is a sneak peak preview of the cover art, seen here for the very first time! What all started, for me personally, as a question, "Why do data science projects fail (80+% of the time)?" turned into a research project, the initial output of which first materialized as a Powerpoint presentation at INFORMS Business Analytics Conference-Houston 2022, then a series of articles posted on INFORMS Analytics Magazine when people said "You should really write all of this down!", and then, finally, a book titled with a statement, Why Data Science Projects Fail, backed up by 200 pages of answers! The idea for the book started when Evan (who hails from Brisbane, Australia---that's right we've never met in person!) saw my articles on LinkedIn, reached out, and said "I think between the two of us we have a good head start on enough material for a book!" By combining real-world stories and lessons learned from our own respective careers in corporations, consulting, and academia, backed up by comprehensive, in-depth research, data, and analysis, Evan and I have crafted a thesis that explains, in detail, why and how data science projects fail most often. The book is intended for practitioners and leaders seeking to improve data science project success rates, resulting in greater business value and economic impact, i.e., which is all of us! The book is also intended for university professors and their students who are jointly engaged in Capstone, Practicum, and Internship courses seeking to go beyond technical subjects (e.g., math, models, and coding) and gain a perspective and a deeper understanding of the more nuanced complexities that so often impair real-world projects. Special thanks to Randi Slack our publisher at Taylor & Francis, Kara Tucker our editor on both the book and my INFORMS Analytics Magazine article series, and Tom Davenport, author of Competing on Analytics and All in on AI, who graciously wrote the Foreword. Stay tuned for more information on release dates and availability wherever books are sold!

96

27 Comments

Here is an interesting read on correlation. I'm particularly keen on how the writer provides a simplified (gentle) summary of an otherwise not so gentle research paper. https://lnkd.in/gMXk5sxS

3

Unlocking the Power of Data: Why Governance Matters in National Statistical Systems 🌐 In today's data-driven world, effective governance is the bedrock of reliable, transparent, and actionable insights. The IAOS-ISI 2024 conference in Mexico explores the critical role of governance in our national statistical systems. Together, with Dominik Rozkrut, Gabriel Gamez and participants of the conference we delved into how strong governance frameworks foster trust, ensure accuracy, and drive informed decision-making. Ready to unleash the full potential of data? Let's connect and elevate our understanding of statistical governance! #DataGovernance #StatisticalSystems #DecisionMaking #DataDrivenInsights"

41

Feeling Lost in Your Data Zoo? Data is powerful, but wading through it can feel like navigating a chaotic animal kingdom! Before you get trampled by information overload, check out part 1 of my new blog, "The Statistical Safari: Exploring Key Concepts (Part 1: Descriptive Statistics)". Ready to unlock the secrets of your data? Head over to my blog and join the data detective adventure! https://lnkd.in/gJ6e-gBD #dataanalysis #statistics #datascience #descriptivestatistics #machinelearning 

7

🤔 How many times have you reinvented the wheel in Tableau?   I thought the radial jittering in my IronViz qualifier was pretty cool.   But it looks like my IronViz mentor, Samuel Parsons, had done the same thing 5 years prior.   At least I settled on the same chart name as he did, so that's what it is - a radial jitter.   These are good for infographics showing a distribution that concentrates around a central point.   📝 In the near future, I hope to make a simple dashboard showing three variations of this chart type, including evenly distributed dots as well as dots on the periphery.   In this #makeovermonday case, I thought the idea of concentrated dots meshed well with firearms - I got the idea from seeing the bee swarms by Shangruff Raina and Sherzodbek I.   Anyhow, I adjusted the worksheet size so that the output is slightly elliptical to fit well in the target image.   ⚖️ I thought I'd go down the route of comparison to domestic purchases because I knew those numbers were high. They've been higher than 2022, but I used that year because it was the last year of the other range and I wanted something semi-current.   🏗→🧠 Build to Learn!   💭🚶‍♀️🚶‍♂️ Follow for more.   #makeovermonday #tableau #data #analytics #VizoftheRay

20

14 Comments

A great review of mine and Adnan Masood, PhD. book ‘Responsible AI in the Enterprise’. This is all becoming increasingly relevant as AI regulation becomes clearer globally and Enterprises are increasingly shaping up how they risk assess and govern AI. ‘Responsible AI in the Enterprise by Adnan Masood, PhD. and Heather Dawe was a great treatment of all the things that can go wrong in AI / ML projects, important principles in the design and operation of these systems, and ways of detecting and mitigating fairness issues in AI Systems.’ Thank you Matt Eland! https://lnkd.in/eWefyccD #responsibleAI #AIEthics #AI

14