Issi Romem

New in CODATA Data Science Journal: Mikala R. Narlock, Scout Calvert, Shawna Taylor, Rachel Priesman Márquez, Aundria Parkman Title: Knowledge Infrastructures Are Growing Up: The Case for Institutional (Data) Repositories 10 Years After the Holdren Memo https://lnkd.in/e6hRxjKC #codata #FAIRdata #OpenScience #datascience #opendata #datasciencej

2

James Perrotta, Tobias Kordsmeyer-Storp, and I recently published a paper in Organizational Research Methods titled “Taking it Easy: Off-the-Shelf Versus Fine-Tuned Supervised Modeling of Performance Appraisal Text”: https://lnkd.in/gDc2P4ur In it, we discuss the practice of scoring psychological constructs/phenomenon from text (e.g., personality traits, positive sentiment regarding a job). Before recently, researchers had to develop their own natural language processing (NLP) algorithms specifically for this purpose, a task that requires moderate to large sample sizes and labeled target scores – the likes of which aren’t always easy to obtain. With the advent of large language models (LLMs), researchers/practitioners can now leverage zero-shot scoring (i.e., “off-the-shelf”) by simply asking the LLM to score the text. However, while easy to do, there are no guarantees that the derived scores will be psychometrically sound. Across six organizational datasets and thousands of comments, we found that supervised NLP (i.e., customized transformer models) produced highly reliable scores. However, and even though not specifically developed for this purpose, we found that off-the-shelf LLMs performed similarly, though with slightly less favorable psychometric properties. The challenge becomes whether or not users can be confident in using LLMs for bespoke scoring on their own data, given those LLMs will not be accompanied by their own psychometric properties supporting the validity of the derived scores. Thus, we propose a decision framework that discusses different strategies to validating LLM scores, including local validation approaches (more time-consuming) and transportability approaches (more difficult to establish). Please check out the paper in ORM! #Artificial Intelligence, #Natural Language Processing

83

4 Comments

I am thrilled to share that I will be a panelist at the 12th International Monetary Fund Statistical Forum: Measuring the Implications of AI on the Economy, taking place on November 20-21 in Washington DC. Join me on November 20, from 10:30am to 12:00pm ET (16:30-18:00 CET), where I will have the honor of presenting the results of the research conducted by the EY Economic Analysis Team: "Macroeconomic Consequences of Generative AI: a Sectoral and Regional Perspective." Our study explores the potential impacts of AI on various economic factors, including the labor market, investment, productivity, GDP, inflation, interest rates, sectoral performance, and international competitiveness, with a focus on Europe, the Middle East, and Africa (EMEA). Drawing upon Acemoglu (2024) and parallels to the ICT revolution, we estimate the share of tasks likely to be automated due to AI over the next decade, along with the consequent gains to ICT investment and total factor productivity (TFP). Utilizing a large-scale global macroeconometric model and a panel error correction model, we project AI’s effects on GDP, inflation, and interest rates. Finally, we employ a recursive-dynamic Computable General Equilibrium (CGE) model to examine AI’s influence on sectoral performance and international competitiveness. I look forward to engaging in insightful discussions and sharing our findings with fellow experts and enthusiasts in the field. See you there! #IMFStatisticalForum #AI #Economy #EY #Research #Macroeconomics #GenerativeAI #EMEA #ICT #Productivity #Investment #LaborMarket #GDP #Inflation #InterestRates #SectoralPerformance #InternationalCompetitiveness #CGE Beatriz Sanz Saiz Raj Sharma Sugan D Palanee Oliwia Komada, PhD Maciej Stefański Piotr Dybka, PhD, Mateusz Maksymiuk Dominika Sikora @Maciej Łopusiński Julia Kotowska Gregory Daco Lydia Boussour Benjamin Falk Sree Sunku

71

3 Comments

Most liberal and conservative colleges Many of you have likely wondered which colleges lean liberal and which skew conservative, but finding a reliable metric has always been challenging—until now! We've developed a robust measure, based on which colleges people graduated from, using data from People Data Labs (https://lnkd.in/gZXKaPxW) and L2 (https://l2-data.com/). Take a look at the chart below for selected colleges. Does it align with your expectations? Any surprises? Thanks Alexia Olaizola and Julian Hayes. #colleges #politics #highered

73

18 Comments

Next week is a big event for labor market data nerds: the Bureau of Labor Statistics releases its preliminary benchmark estimate for March 2024 - a big component of eventual benchmark revisions to key BLS data. I wrote a too-long preview of that event on my substack: https://lnkd.in/eTvWT8Sk Key takeaways: 1/ The benchmark revision is likely to be negative, enough to shift our assessment of job growth during the March 2023 - March 2024 period from "strong" to "solid". But it won't look recessionary. 2/ This is "old news". We have much fresher, high quality data that's giving us a good picture of the labor market trajectory. Data from 4.5-16.5 months ago won't alter that. But I've spoken to enough savvy people who think that to financial markets or "the vibes", old news can matter. 3/ This revision won't affect two of our most important labor market data sources: key household survey ratios like the unemployment rate and the prime working age employment/population ratio; and data on unemployment insurance.

47

1 Comment

Quick reminder - on Wednesday we get the preliminary benchmark estimate from the BLS, which is likely to be negative. My explainer is below:

5

I don’t know about all of you, but my life is chock full of examples of “never say never.” I said I would never drive a mini-van, and yet I ended up doing so for nearly 15 years. I said I would never watch reality TV, and yet my wife and kids have sucked me into watching all kinds of things with them from Project Runway to The Great British Baking Show. I said I had no interest in ever doing a triathlon and now I am planning on trying one this summer (a little one 😶). So I should have realized when I said I would never write a book, I suppose it was inevitable that I would, indeed write a book.    I never thought I would write a book because (1) it seemed extremely daunting and (2) I just didn’t think I had “book-length” thoughts about any particular subject. But as CEO of NORC at the University of Chicago, I have spent the last decade making the case for using trustworthy and reliable data and analysis to inform society’s most important decisions. During this time, we’ve seen the actual concept of objective data and analysis increasingly questioned, the proliferation of poorly constructed data and sloppy analysis, along with data and analysis that are designed from the outset to distort the truth. It became apparent that organizations like NORC, and people like me, have an obligation to do more to help explain why trustworthy data and analysis are central to understanding who we are and where we are going. So these were the reasons I wrote Fact Forward… which you can learn a little more about below.

90

16 Comments

MIT professor Catherine D'Ignazio's new book “Counting Feminicide” highlights the work of data activists who use data to fight for human rights and dignity. These inspiring stories showcase how data collection, sharing, and analysis can be powerful tools for positive social change. #dataactivism #socialchange #MIT https://lnkd.in/eQQJvWfp

25

New FET article: A legacy of bad economics from Adam Smith’s Pin Factory Did Smith learn the wrong lesson from his factory tours? And why do we keep reciting it uncritically? Please like, share and subscribe. https://lnkd.in/g5Sq8GCu

6

3 Comments

Which is more important – understanding the flight of the arrow or pinpointing where it will land? A trade-off exists when using machine learning for forecasting #ThankFathomitsFriday https://lnkd.in/ehVdYyhg #TFiF #Machinelearning #ML #Economics #DataScience

1

Inspired by Ken Flerlage's 2020 Tableau dashboard on different ways to visualize election data, I built something similar using state-level 2024 presidential election results in Flourish. Check out all of the 15 different approaches here: https://lnkd.in/eCWvqG4z

18

🤔 Can complex models improve performance? As the accuracy vs. interpretability debate reveals, sometimes simpler models are just as effective, especially when dealing with structured data. Read Conor O'Sullivan's latest article now! #MachineLearning #DataAnalysis

12

The working-age population growth has returned to positive figures after a period of stagnation. This resurgence is largely attributed to increased border apprehensions and subsequent releases into the U.S., fueling substantial population growth. It's important to note, however, that this chart significantly underestimates the actual increase in population. Official government statistics do not fully capture the extensive wave of immigration. Recent trends, however, suggest a potential reversal in this pattern. Latest data reveals a significant decrease in border apprehensions, a trend that might persist following the administration's executive order issued last week, which aims to further reduce these figures. If the immigration wave dies down, working-age population growth is likely to stagnate like it did before the pandemic. The aging of the baby boomer generation into retirement age is leading to a natural deceleration in the growth of the working-age population. I have utilized the 12-month moving median of the monthly growth rate in the working-age population, which helps smooth out the series breaks that typically occur every January. #labormarkets #demographics #population #laborforce #recruitment

44

1 Comment

TL;DR: this week's jump in initial filings for unemployment insurance ("jobless claims") probably reflects noise in the data rather than a fundamental increase in layoffs. I usually don't bother writing up the weekly jobless claims data on here, especially because they haven't shown anything interesting recently. (Aside from the recurring theme of "despite scary headlines, layoffs remain unusually low and stable by historical standards.") That said today was "interesting", because initial filings spiked up to ~230K vs the recent average of around 210K. This isn't a very high number, but it was the highest since August 2023. If it sticks would probably be indicative of a modest increase in layoffs (and would probably be enough to gradually and slightly increase the pool of unemployed people). That said... about half the national increase came from a sudden spike in New York state, and in particular NY state's transportation & warehousing industry. There have been no indications of mass layoffs in NY. Instead, one possible culprit is a calendar shift in spring breaks - leading school bus drivers who work for private contractors to file for unemployment insurance. We'll see, of course, but I doubt it indicates a sustained pickup in layoffs.

19

1 Comment

Social network analysis is now a core component of computational social science. D-Lab Data Science Fellow Christian Caballero introduces concepts of network analysis in this week's D-Lab Blog Post. Check out his post to gain a foundational understanding of networks and how they can be used for social science research: https://lnkd.in/g5vk9uks

5

While overall #inflation continues to decline, #childcare prices continue to climb. • The August data from the Consumer Price Index (#CPI) show that overall prices increased 2.5% year over year; the daycare and preschool subcomponent surged 6.2% • This is the largest increase since May 2023 and shows how the childcare crisis is continuing to get worse • As inflation took off in the US, overall inflation outpaced daycare and preschool from March 2021 to February 2023 (see chart below); but this is the exception historically… • Since the #BLS began collecting data about daycare and preschool in 1991, the % change has been larger than overall inflation in 327 out of 393 months, or 83% of the time

18

Ever wondered how federal data on race and ethnicity evolves over time? Read as Jonathan Schwabish explores the shifting landscape of Census categories and the public's role in shaping this critical data. https://lnkd.in/eNZ4Yc9r #DataViz #DataVisualization #CensusData #DataAnalysis

16

1 Comment