this week i

attended r conference in the land of eusilc.  data anonymization for open science course focused on statistical disclosure control: synthpop, simpop, generative adversarial networks, k-anonyminity: if, for each combination of values of quasi-identifiers, at least k records exist in the dataset sharing that combination.  perturbation methods: noise masking (adding normally-distributed errors), microaggregation (grouping similar records, replacing w means or medians), record-swapping, rounding, resampling, pram (probabalistically altered categorical data).  puf techniques to thwart evildoers

longitudinal data are the hardest type to anonymize

the layout of hypercubes are codified in european legislation

preserve the correlation structure

simulated annealing

 

keynoted: cran reached 25th birthday with 80 submissions daily, r core commit stats log-scaled due to brian ripley, abhishek agreed to debug 2009-me

maybe the code is fine but the world changed 

r not zero indexed but you europeans and your 1st floor is the zero-th floor

r is reading your code as a parsed tree

"commit a bunch of files before lunch 🍝"




dislike unswingable chandeliers. "the square bracket method is just ginormous" "males have this accident bump btw 18-25 when they do stupid things"
 

thanked hadley wickham for session scraping forbes, useful to add 400 richest to scfi recognize your name from twotorials  (in-class notes below)


 

scf + forbes 400
 
wealth tax rates are, respectively, 2.5 and 4.5% under the warren and sanders plans
 


read preferred reporting items for complex sample survey analysis (pricssa) by seidenberg, moser, & west.  every complex survey needs table one

we propose .. an itemized checklist
 
a review by briesacher et al. (2012) found that less than half of papers analyzing data from the medicare current beneficiary survey described appropriate weighting or variance estimation
 



democracy is not a sufficient condition for redistribution
 
top income shares in brazil have followed a sine wave pattern
 
i tracked down data for 71 years between 1926 and 2015.  the tabulations are generally fine-grained, consistent, and comparable by international standards
 
the data comes from the decennial censuses.  figures for intercensal years were computed via cubic spline interpolation
 
brazilian tax law never defined the tax unit very clearly
 
international comparisons should always be interpreted cum grano salis

 

read the frequently asked questions section of metaculus.  demography ought to be bigger, featured tutorial obsolete points

we follow a few principles to elevate forecasting above simple guesswork

greater weight to predictions by forecasters with better track records

a good question will be unambiguously resolvable 

 

read income inequality in the united states: using tax data to measure long-term trends by gerald auten and david splinter.  valid critique of piketty but focused on surrogate outcome: income < consumption < wealth inequality.  dragon lair treasure mounds swell both by acquisition, appreciation

marriage rates on tax returns declined from 67 to 37 percent between 1960 and 2019.  however, marriage rates have remained high among the top one percent, decreasing only from 90 to 85 percent.  declining marriage rates outside the top of the income distribution increases income shares at the top of the distribution.  larrimore (2014) estimated that the differential decline of marriage rates explains 23 percent of the increase in household income gini coefficients between 1979 and 2007

 note that top wealth shares ranked by wealth are higher than when ranked by income

 

 

in the 1960s, only a tiny fraction of taxpayers actually paid the top tax rates (fewer than five hundred tax returns in 1962), in part due to tax avoidance behavior

the kakwani index of tax progressivity summarizes average tax rates over the entire income distribution.  while changing little between 1962 and 1985, this index increased dramatically from 0.07 to 0.29 between 1985 and 2019

the u.s. tax system is more progressive than in european countries, which rely more on regressive value-added and payroll taxes

our estimates for after-tax income indicate that the top one percent share increased only 1.4 percentage points since 1979 and only 0.2 percentage points since 1962 .. using only market income on tax returns, piketty and saez (2003) argued that the top one percent share of income more than doubled since 1962.  this analysis, however, did not include transfers and other income sources not reported on individual income tax returns, nor did it account for the effects of major tax reforms and changes in marriage rates.  thus, it gave a distorted view of income inequality levels and trends


7/25