The Curious Engineer

All blogs / Math

Gojiberries

gojiberries.io

Science, economics, statistics

MathSociety / politics / economyAI / ML

’05’10’15’20’25
Posts
491 from 26 Dec 2004 to 1 Oct 2026
Posting
weekly (30 in the last year)
Kind
personal blog
Feed
https://www.gojiberries.io/rss/

2026 23

  1. Do Spammers Deserve Death?
  2. Cache-Control for LLMs
  3. Reverse Dependency Checks for Empirical Outputs
  4. Notes from a fan
  5. Science: 20 GPs in a single District
  6. Ceci N'est Pas un Proxy
  7. Fifty Meters Apart
  8. The Ties That Bound
  9. Join, or Die
  10. Ceci N'est Pas une École
  11. Recommended Treatment
  12. No Logbook, No Problem: Recovering LATE When You Can't Observe Compliance
  13. Epic Children, Mostly Sons
  14. Regression to the Mean Model
  15. The Other Gender Gap
  16. I-Ran the Numbers
  17. Comparative Advantage: Calibrated Pairwise Win Rates from LLM Judge Scores
  18. Liberation and Immanence
  19. The Politics of Struggling Americans
  20. Anekantavad
  21. Hardware‑Attested Ad Insertion for Encrypted Video
  22. Out-of-Context: Constrained Tool Based Exploration of Context
  23. The Second-Best of Specific Worlds

2025 45

  1. First, Assume No Elections
  2. Targeted Model Editing
  3. Strings Attached: Shipping Fuzzy‑Joined Data
  4. Pity The Rich
  5. Chaotic Flows: LLMs in Low‑Tolerance Workflows
  6. PK
  7. Stable LLM Inference
  8. Identifying Assumptions
  9. What Problem Is Traditional RAG Solving?
  10. Conscious Decoupling: Separating Budgets For Thought and Action
  11. Testing LLMs: Engineering Confidence Without Certainty
  12. Simulating and Evaluating Agentic Systems
  13. What Problem is DSPy Solving
  14. The Case For Pay-as-Bid GPU Pricing
  15. The Double Descent of Standard Errors
  16. Testing Distributional Implications of LATE
  17. Unbiased regression with costly item labels
  18. A Program of Experiments With Risk, Context, and Organizational Constraints
  19. The Elephant in the Sampling Frame: Lessons from Basu's Parable
  20. How to Make Things Slower So They Go Faster
  21. Bandwidth Selection Without Grid Search
  22. Behavioral Validity Checks for ML‑Based Coding
  23. Rank‑Preserving Calibration for Multiclass Probabilities
  24. Generalization Gap in Over‑Parameterized Models
  25. Burning ₹168 to Earn ₹100
  26. A Lightweight ALS Solver for Iterative GLS
  27. One Record, Many Assertions
  28. The Example Store Is Model State
  29. npm fund
  30. Advertising Without Signal: The Rise of the Grifter Equilibrium
  31. Boosting Stability: Reducing Instability Under Row Permutation
  32. From Vibe Coding Assistants to Coding Assistants
  33. A Pink Revolution in Indian Policymaking
  34. WAR For Cricket
  35. (Don't) Forget About It: Toward Pareto Improving GD
  36. Greedy is Good. Less Greedy May be Better.
  37. Hungary For More? Optimal 1-to-Many Matching for Causal Inference
  38. Oracles, Not Just Scientists. The Seemingly Exceptional Batting Average of Pre-Registered Hypotheses
  39. Problem Solving
  40. PageRank Measures the Wrong Fixed Point Under Polarization
  41. The Undersupply Theorem Meets the Extraction Equilibrium
  42. Bias as a Fix for Congestion
  43. Resolution Lost: The Deadweight Costs of Strict Isotonicity
  44. Picks, Shovels, Superstars, and Poachers: Explaining the AI Investment Boom
  45. Using Reward Models to Select Pre-Training Data

2024 8

  1. What do we learn from bestseller regressions?
  2. Political Psychology: What Political Scientists Get Wrong About LaPiere's 1934 Study
  3. Unbiased Hiring
  4. Not so Prompt: Prompt Optimization as Model Selection
  5. The Welfare Costs of Low-Friction Idea Production
  6. Fun Facts
  7. Less Known Facts About Muslims in India
  8. The Relative Status of Brahmins Across India in 1931

2023 6

  1. A Benchmark For Benchmarks
  2. Inter-Group Prejudice
  3. Time Will Tell
  4. Limits of Harms From Affirmative Action
  5. Not Normal, Optimal
  6. When Is Discrimination Profit-Maximizing?

2022 6

  1. Tipping Lemons: Market Failures in Tips
  2. Vote Geography, Wasted Votes, and AIMIM’s Seat Bonus
  3. Back to the Future: Engineering Without Data (Models)
  4. Noise: A Flaw in Book Writing
  5. Make Government Code Publicly Available
  6. Sampling Domain Knowledge

2021 6

  1. Pastations: Is it Time to Move Beyond Presentations?
  2. The Nonscience of Machine Learning
  3. The Story of Science: Storytelling Bias in Science
  4. The Hateful ATE: The Effect of Affective Polarization
  5. To a Better End: How the Middle Can Improve the End
  6. This Time It's Different: Polarization of the American Polity

2020 7

  1. Liberalizing Daughters: Do Daughters Cause MCs to be Slightly More Liberal on Women's Issues?
  2. Sense and Selection
  3. Another ANES Goof-em-up: VCF0731
  4. Survey Experiments With Truth
  5. Routine Maintenance: How to Build Habits
  6. Gaming Measurement: Using Economic Games to Measure Discrimination
  7. Predicting Reliable Respondents

2019 2

  1. Snakes on Ladders: Encouraging People to Climb the Engagement Ladder
  2. The Risk of Misunderstanding Risk

2018 3

  1. Growth Funnels
  2. Technical Leadership: Building Great Teams and Products
  3. Self-Redaction

2017 10

  1. Bad Hombres: Bad People on the Other Side
  2. Taking a Holiday from Looking for Jobs
  3. Party Time
  4. Estimating Bias and Error in Perceptions of Group Composition
  5. Searching for Mr. President
  6. God, Weather, and News vs. Porn
  7. How Do We Know?
  8. (Software) Product Development Cycle
  9. The Innumerate American
  10. Motivated Citations

2016 2

  1. Town Level Data on Cable Operators and Cable Channels
  2. Ipso Facto: Analysis of Complaints to IPSO

2015 2

  1. Getting a Measure of Measures of Selective Exposure
  2. Where's the Porn? Classifying Porn Domains Using a Calibrated Keyword Classifier

2014 4

  1. Tale of Two Scandals: Impact of Benghazi and IRS Scandals on Obama Approval
  2. Liberal Politicians are Referred to More Often in News
  3. Cable Ownership by DMA by Race Over the Years
  4. Error Free Multi-dimensional Thinking

2013 1

  1. Not Feeling as Warm toward Whites Anymore

2012 2

  1. Bias is in the Eyes of the Beholder
  2. State of the Union

2011 6

  1. Correlation Between Social Welfare and Cultural Preferences by Party
  2. Nudging
  3. On (Modest) Differences In Racial Distribution of Voting Eligible Population and Registered Voters in California
  4. Creating A Leaky Internet
  5. Fairly Random
  6. Surveys in Exchange for Temporary Internet Access

2010 8

  1. Delay Shooting the Help
  2. 'Automatically' Summarizing Articles and Judicial Opinions
  3. Monetizing Websites using Mechanical Turk
  4. Personal Milestone Campaigners
  5. Lifting the Veil on Some Issues Around The Burka Debate
  6. Opening Interfaces to Websites and Software
  7. An Appetizing Question: Why Don't We Read Nutrition Labels?
  8. Internet Panel + Media Monitoring

2009 2

  1. Prospect Theory in Retrospect
  2. The Problem of Pakistan

2008 8

  1. On Representation in a Democracy
  2. News Coverage of Sarah Palin
  3. Populism at the Gates
  4. The Art and Artifice of Frida Kahlo
  5. Military Experience of US Presidents
  6. The Case for Epistemic Controls in a Democracy
  7. The Hubris of Journalists
  8. Reducing the Nuclear Threat by Reducing Local Threat Perceptions

2007 9

  1. Vague Apprehensions
  2. Reporting Politicking: Horse-Race Coverage of Politics
  3. Movie Review: The Namesake
  4. Interpretive Approaches in Art History
  5. Secret Loss in 'Peach Blossom Land'
  6. Democracy: Whither Epistemic Validity?
  7. Why Not to Think Like a Lawyer
  8. Conversation with Glenn Frankel: Principles, Ethics, and the Practice of Journalism
  9. Conversation With Glenn Frankel: Principles of Good Journalism

2006 4

  1. A Small Government? US Federal Budget as a Proportion of the Economy
  2. What is so Foreign About Foreign Aid?
  3. Understanding Voter Disinterest
  4. Playing with Numbers

2004 1

  1. Some Die Young

Undated 326

  1. When ols and sur disagree
  2. Streaming calibration
  3. Good e
  4. Data police
  5. I endorse this message learning from endorsement experiments
  6. Better split decisions more stable cart by using subsampled node level cv
  7. Measuring tree similarity
  8. Pareto deployments
  9. Whats wrong adversarial llm judges with evaluation criteria
  10. Dismissed without prejudice evaluating prejudice reduction research
  11. Better starts smarter clusters lookahead initialization for k means
  12. Ditty for bush
  13. The invisible hand needs a hand two short stories on market design
  14. The modal social science dataset is under analyzed
  15. Credit where it is due two facts about credit cards
  16. Comparing ad targeting regimes
  17. Diabetes as a dummy variable
  18. A textbook paradox the puzzle of price dispersion in e books
  19. Internet search is not a naive information retrieval problem
  20. Against complacency
  21. How numerous are the numerate
  22. Literacy in india can india spell success
  23. Predicting failure using ml for measurement
  24. Feigning competence checklists for data science
  25. A brief history of presidential debates
  26. Need for epistemic standards for evidence and argumentation in governing bodies
  27. Siamese networks for record linkage
  28. Journalists and numbers
  29. Impact of menu on choices choosing what you want or deciding what you should want
  30. Trump trumps all coverage of presidents on network television news
  31. Code 44 how to read ahler and sood
  32. Insanity that is us military expenditure
  33. Optimal persuasion
  34. Confirmation bias
  35. What clicks with the user maximizing ctr
  36. Attention please dealing with inattentive respondents in experiments
  37. Optimal data collection
  38. How many people does it take
  39. Homing in on the home advantage
  40. Limits of social science
  41. Hard problems about research software
  42. When cost of misclassification is much higher for the customer than the business
  43. Musings
  44. Peer review
  45. From lost lives to life lost
  46. Companies with benefits or potential welfare losses of benefits
  47. 94 5 certain that covid vaccine efficacy will be less than 94 5
  48. Lawyers
  49. Beyond anomaly detection supervised learning from bad transactions
  50. Matching dirty data
  51. Making an impression learning from google ads
  52. Trading on overconfidence
  53. Bad service missing functionality in web services
  54. Smallest loss that compute can buy
  55. Measuring segregation
  56. Organic labels better labels from organic growth
  57. Loss of confidence
  58. Optimal sequence in which to schedule appointments
  59. Against proxy variables
  60. Talking on a tangent
  61. Hello world
  62. Son bias in the us evidence from business names
  63. Interview with bill thompson part iii copyright law
  64. Clustering and then classifying improving prediction for crudely labeled and mislabeled data
  65. Estimating hillarys missing emails
  66. Bijli sadak makaan art at the crossroads of infrastructure and culture
  67. Just browsing internet surveys and old broswers
  68. Missing daughters of indian politicians
  69. Good nyt provision of not news in the nyt over time
  70. Countpy incentivizing more and better software
  71. Fair game
  72. Mean ideology of state congressional delegation over the years
  73. Measuring partisan affect coldly
  74. Gandhi and his critics
  75. Interview with bina shah part 2
  76. Sampling with relative primes
  77. Incentives to care
  78. Unlisted false negatives
  79. What is so middle about middle class
  80. Why were polls so accurate
  81. Estimating trend at a particular point in a noisy time series
  82. The misinformation age provenance is not enough
  83. Learning from the future with fixed effects
  84. Too much churn
  85. Monetizing bad models pay per correct prediction
  86. Not recommended why current content recommendation systems fail us
  87. The origins of personalization
  88. Big failure too complicated to communicate
  89. Republican census
  90. Far beyond the narrow database
  91. Dont expose yourself discretionary exposure to political information
  92. Nothing to see here
  93. Strapped for time
  94. Fooled by randomness
  95. Optimal targeting 101
  96. Reducing frictions in selling data products protecting ip and data
  97. Afib apple watch did not cause increase in atrial fibrillation diagnoses
  98. Optimal recruitment for experiments using pair wise matching distance to guide recruitment
  99. Nextdoor
  100. Sampling on m turk
  101. Interviewer assesments of respondents level of political information
  102. Moving further away from the out party
  103. Does gender of a child cause partisanship
  104. French for bread
  105. The misinformation age measuring and improving digital literacy
  106. Partisan morality
  107. The technology giveth the technology taketh
  108. The unscientific republican
  109. Automating understanding not just ml
  110. The human in the machine semi automated approaches to ml
  111. Some facts indian polling stations
  112. Steming the rot relative deprivation
  113. Canonical insights
  114. Why do people re elect bad leaders
  115. 85 nonsense predicting 85 percent of cyber attacks using input from human experts
  116. Campaigns construction and moviemaking
  117. Online learning with biased sampling
  118. Interacting with human decisions heterogeneous treatment effects with human decisions as covariates
  119. Teaching social science
  120. Rent seeking why it is better to rent than buy books
  121. Pride and prejudice
  122. Operational efficiency the road to efficient operations
  123. Feelings towards groups
  124. Numbers
  125. Computing optimal cut offs
  126. Market welfare
  127. Amartya sen
  128. Firmly against posing firmly
  129. Network targeting
  130. Explore exclusive exploitation the billboard top 100 method of learning
  131. Generative ai and the market for creators
  132. Interpreting data
  133. Is price right dramatic variability in prices of generic goods over short periods on major web retail platforms
  134. No comparative advantage
  135. Not so robust the limitations of doubly robust ate estimators
  136. Wanted effects that support my hypothesis
  137. Bayes and representativeness heuristic
  138. Autosum plus
  139. Thats smart what we mean by smartness and what we should
  140. Missing women on the streets of delhi
  141. https://www.gojiberries.io/6000/
  142. The value of predicting bad things
  143. Solving problem solving meta skills for problem solving
  144. Conceptualizing effect of a deliberative poll
  145. The other side
  146. Partisan retrospection
  147. Living with history
  148. A b testing recommendation systems
  149. Best guess of true proportion of 1s
  150. Life expectancy in the war zone gendered perspective
  151. I want the same capuchin monkeys and fairness
  152. Towards a two sided market on uber
  153. Stereotypical understanding
  154. Randomly sampling from the streets
  155. Reducing errors in survey analysis
  156. The new unit of scientific production
  157. Ruined by google
  158. American pii lapses in securing confidential data
  159. The value of money learning from experiments offering money for correct answers
  160. Decoupling compliance and experimental designs
  161. Executive capture rent seeking within companies
  162. The declining value of interpersonal advice
  163. Why do we fail and what to do about it
  164. Quitting at 40
  165. Everywhere meeting consumers where they are
  166. The best we can do is responsibly answer the questions that life asks of us
  167. Sometimes scientists spread misinformation
  168. Deliberation as tautology
  169. Random redistricting
  170. Quality data plumbing ml data pipelines
  171. Disgusting
  172. Whats best comparing model outputs
  173. Stemming link rot
  174. Advice that works
  175. Raising money for causes
  176. The mystery of price dispersion on amazon
  177. Instrumental music when it rains it pours
  178. What is behind the reproducibility crisis
  179. Unmatched the problem with comparing matching methods
  180. Citing working papers
  181. Subscribing to unpopular opinion
  182. Ruling out explanations
  183. Expertise as a service
  184. Undercited findings
  185. What academics can learn from industry
  186. Equilibrium fairness
  187. Ml oops what data to collect part 3
  188. Ml oops improving and deploying on device models with confidence part 1
  189. Ml oops keeping track of changes part 2
  190. 99 problems how to solve problems with data
  191. Compensation with currency with no agreed upon value
  192. Fairly certain understanding fairness in terms of uncertainty in prediction
  193. It pays to search
  194. Contingent answers effect of quotas on womens representation
  195. Optimally suboptimal behavioral economic product features
  196. No stopping impact of the stopping rule on the sex ratio
  197. Traveling salesman
  198. Political macroeconomics
  199. Cell mates extracting useful information from tables for llms
  200. Building together separately challenges of software development
  201. Superhuman level performance
  202. Interview with bill thompson part iv fragmented information
  203. Interview with glenn frankel part 5 foreign reporting and technology
  204. Interview with glenn frankel part 1 biographical focus
  205. Interview with glenn frankel part 2 early professional experiences
  206. Interview with glenn frankel part 4 reporting on emotive issues
  207. Interview with bill thompson part ii political economy of internet
  208. Interview with bill thompson technology columnist for the bbc part i
  209. The missing political economy of free money
  210. Social science in a guest appearance in the play capitalism
  211. Building above average software software for the lay user
  212. Institutional distrust
  213. Sigh tations
  214. Mass affective polarization
  215. No shit open defecation in india
  216. Scan to save arbitrage between paper and digital
  217. Uncorrelated tariffs
  218. Dont feed the robots
  219. The github sampler
  220. Results may vary
  221. The paradoxes of condorcets jury theorem when applied to democracy
  222. Analyzing epistemic gains and robustness of prediction markets
  223. Why elections matter
  224. Extremely or fairly satisfied with democracy
  225. Democratic pandering
  226. Difference in coverage
  227. Some facts about politifact
  228. India in 1000 words or less part 1
  229. On love
  230. The problem in immigration reform
  231. Complimentary spam
  232. Sex vs jihad
  233. Sidhwas lahore a lovingly embroidered family heirloom
  234. Interview with bapsi sidhwa
  235. Interview with bina shah
  236. Sorted or polarized
  237. Rasmussen vs the rest
  238. Why are prices the same
  239. Missing market for academic experts
  240. Striking changes among democrats on race and gender
  241. Coding at a distance
  242. Aging curve of political knowledge
  243. Take 80 mg atoravastatin for myalgia
  244. The fairest of them all
  245. Optimal sequence in which to service orders
  246. Reviewing the peer review with reviews as data
  247. Q tips for r
  248. R query matrix indexing
  249. Carving up the academic pie how are academic disciplines divided
  250. Waxing poetic
  251. Perils of balancing scales
  252. Star trek trekking uncertainly between utopia and earth
  253. Musharraf by numbers corruption
  254. The generals report card education under musharraf
  255. Interview with saira wasim
  256. Reading readers needs
  257. R recoding variables reliably and systematically
  258. Comparing datasets reporting only non duplicated rows
  259. The missing plot
  260. Searching for great conversations
  261. Topgraphy of book sales what lurks beneath
  262. Bad weather getting weather data by zip and date
  263. A quick scan from paper to digital data
  264. Enabling foia
  265. The benefit of targeting
  266. Orwell times corporate beneficence
  267. 5 is greater than 1 9
  268. Interview with eric berlin executive producer of bc
  269. Interview with dr victor stenger
  270. From satellites to streets
  271. Comments please the future of blog comments
  272. Piracy the real mantra behind indian success and what the wall street journal doesnt get
  273. Disproving god
  274. Wheres the news classifying news domains
  275. Working with modestly large datasets in r
  276. Merchants of art
  277. Profit maximizing staffing
  278. India in 1000 words or less e2 80 93 part 3
  279. Proposal for a small ngo fomenting cross class interaction
  280. Causality and generalization in qualitative and quantitative methods a second look part 1
  281. The myth of knowledge
  282. All politics is identity politics
  283. Small warning about using anes cumulative file
  284. No props for prop 13
  285. Cricket an unfairly random game
  286. Re imagining online learning
  287. Whats the next best thing to learn
  288. Fat or not proper training of dl models
  289. The value of bad models
  290. Interview with infoq
  291. Rape in india
  292. Gss and anes a pattern of difference
  293. Where the law stands and beyond first amendment and lobbying
  294. Spincycle supports obama
  295. The theory of all
  296. Understanding delhi and e2 80 98the delhi walla e2 80 99
  297. Did indias economic miracle begin in 1980 why
  298. Clarifaing the future of clarifai some thoughts
  299. I recommend it a recommender system for scholarship discovery
  300. The order of feelings
  301. A potential source of bias in estimating impact of televised campaign ads
  302. Sample this
  303. The worry about anna hazare
  304. Weighting to multiple datasets
  305. Size matters significantly
  306. Estimating impact of sample representativeness in experiments where participants interact
  307. Postively polarized some potential negatives of elite polarization
  308. Learning about delhi a review of the delhi walla books
  309. Comment flawed analyses in deliberative polls
  310. A walk down the past connaught place
  311. Coding issues in the anes cumulative file
  312. Political economy of everyday conversation
  313. Selected ethnography of marketing in india
  314. News and etcetra
  315. Desideratum qualitative vs quantitative methods
  316. Role of argument in sciences law and social sciences
  317. India in 1000 words or less e2 80 93 part 6
  318. End of information hierarchy
  319. Propagation of bias
  320. Education and economic inequality
  321. Correcting for differential measurement error in experiments
  322. Congenial invention and the economy of everyday political conversation
  323. Interpeting clusters and outliers from clustering algorithms
  324. No value added models
  325. When people treat each other
  326. Remembering reagan