Slides from Some Talks

For additional talks, browse the directory or email me.

UIUC 2025 | MEG 2024 | CEBA 2024 | MEG 2023 | TAMU 2022 | MEG 2022 | Stata 2021 | ES meetings 2021 | UC Santa Cruz 2021 | Chicago 2020 | BU 2020 | Yale 2019 | MEG 2018 | UCONN 2018 poster .pdf .tex .sty | Duke 2018 | NASMES 2017 | MEG 2016 | KSU 2015 | ICDM 2007

In-page Links to Paper Details

Finite-Sample Inference on Auction Bid Distributions
Multiple Testing of a Function's Monotonicity
Ordinal Regression and Decomposition
Multiple Testing of Ordinal Stochastic Monotonicity

Inference on Consensus Ranking of Distributions (JBES)
Conditions for extrapolating consumption differences to welfare differences (EI)
Confidence Intervals for Intentionally Biased Estimators (ER)
Comparing latent inequality with ordinal data (EctJ)
Comparing distributions by multiple testing (JoE)
distcomp: Comparing distributions (Stata Journal)
Smoothed estimating equations for IV quantile regression (ET)
(sivqr) Smoothed instrumental variables quantile regression (Stata Journal)
k-Class Instrumental Variables Quantile Regression (Emp Econ)
Cognitive norms for dementia classification (J Neuropsychology)
Frequentist properties of Bayesian inequality tests (JoE)
Smoothed GMM for quantile models (JoE)
Inference on (conditional) quantile differences and interquantile ranges (EctJ)
Fractional order statistic quantile inference (JoE)
Quantile inference by fixed-smoothing asymptotics and Edgeworth expansion (JoE)
High-order Coverage of Smoothed Bayesian Bootstrap Intervals for Population Quantiles (Austrian Journal of Statistics)
A computational approach to style in American poetry (Int'l Conf Data Mining)

Resting/superseded

Grad school work

Working Papers

Finite-Sample Inference on Auction Bid Distributions Using Transaction Prices

2024, submitted
(with Xin Liu)

paper | code/tex/etc. | .bib

Finite-sample, nonparametric, uniform confidence bands for the underlying bid quantile function under symmetric IPV when only the transaction price is observed (i.e., only one order statistic). Extensions: a) varying number of bidders, b) auction-level unobserved heterogeneity (including new bounds). Empirical: assess heterogeneity across number of bidders (wrt "exogenous participation") and observed auction characteristics.

Multiple Testing of a Function's Monotonicity

2024, submitted
(with Wei Zhao)

paper | code/tex/etc. | .bib

Where in its domain is an unknown function increasing? (Ex: structural function, average index function, functional coefficient, conditional quantile, etc.) We develop a multiple testing procedure that can also be inverted into confidence sets.

Ordinal Regression and Decomposition

2024, submitted
(with Qian Wu)

paper | code/tex/etc. | .bib

OLS and Blinder–Oaxaca decomposition with an ordinal outcome variable coded as 1,2,3,…: how to interpret? For OLS, prediction depends on these being true cardinal values, but description does not: the conditional "mean" function can be interpreted as a sum of conditional survival function values. For decomposition, the Blinder–Oaxaca estimated explained proportion is numberically equivalent to a counterfactual-based survival function decomposition using OLS (linear probability model) distribution regression to estimate the counterfactual. That is, the naive 1,2,3,… Blinder–Oaxaca results have a meaningful interpretation robust to any true cardinal values. In our empirical decomposition of rural–urban mental health differences, we describe a nonparametric estimator and compare it to Blinder–Oaxaca, which has the added advantage of easy detailed decompositions.

Multiple Testing of Ordinal Stochastic Monotonicity

2023, submitted
(with Qian Wu)

paper | tex/etc. | .bib

Where is Y "increasing" in X in the distributional sense of stochastic monotonicity? We propose and justify an FWER-controlling multiple testing procedure, which can also be inverted into confidence sets. We apply this to study the relationship between mental health and education.

Publications

Browse the directory of all draft PDFs, replication .zip files, etc.

Inference on Consensus Ranking of Distributions

2024, Journal of Business and Economic Statistics

published | FREE (1st 50) | accepted | code/tex/etc. | more examples | .bib

I propose methods to learn from data about the set of utility functions for which one distribution is preferred over another (higher expected utility). Results are more informative than an all-or-nothing test of unanimous agreement (stochastic dominance). The economic interpretation differs from that of CDF-based restricted stochastic dominance.

Confidence intervals for intentionally biased estimators

2024, Econometric Reviews
(with Xin Liu)

published | FREE (1st 50) | accepted | code/tex/etc. | .bib

Simple CIs using estimators that are intentionally biased to reduce MSE (like sivqr/SEE-IVQR). At 95% level, improves length and coverage compared to unbiased benchmark.

Conditions for extrapolating differences in consumption to differences in welfare

2024, Economic Inquiry
(with Wei Zhao)

published | accepted (embargo till 3may2025) | tex | .bib

Statistical tools can learn about consumption distributions but not directly about the resulting welfare. Even first-order stochastic dominance in consumption does not imply higher welfare, depending how consumption risk is allocated across individuals with different risk preferences. We provide various conditions under which SD1 in consumption indeed does imply higher welfare.

k-Class Instrumental Variables Quantile Regression

2024, Empirical Economics
(with Xin Liu)

published | accepted | code/tex/etc. | .bib

We apply k-class estimation to IVQR to reliably reduce median bias for certain choices of k.

Comparing latent inequality with ordinal data

2023, The Econometrics Journal
(with Wei Zhao)

published | accepted | code/tex/etc. | extra examples | .bib

We provide identification and inference results to compare two latent distributions (better? more dispersed?) when only ordinal data are available, without unrealistic assumptions.

In Stata: issue command
net from https://kaplandm.github.io/stata
and follow instructions for installation of the latentcs command (and email me if you have problems). Note: older Stata versions do not support https, in which case you can download the "code/tex/etc." .zip file above.

The impact of conventional versus robust norming on cognitive characterization and clinical classification of MCI and dementia

2023, Journal of Neuropsychology
(with Alyssa N. Kaser, William Goette, and Andrew M. Kiselica)

published | public/NCBI | .bib

High-order Coverage of Smoothed Bayesian Bootstrap Intervals for Population Quantiles

2023, Austrian Journal of Statistics
(with Lonnie Hofmann)

published (open) | accepted | code/tex/etc. | .bib

Motivating further study in other settings, we provide some results for continuity-corrected Bayesian bootstrap (Banks, 1988) confidence intervals for population quantiles:
(very) high-order accurate · exact in special cases · no smoothing parameter required

Smoothed instrumental variables quantile regression
(sivqr Stata command)

2022, Stata Journal

Adapted in Stata 18 as ivqregress smooth

published | accepted | code/tex/etc. | .bib

I introduce a Stata command implementing Kaplan and Sun (2017)

Update: sivqr now works with qregplot! Thanks to Fernando Rios-Avila; to install/update: ssc install qregplot, replace

In Stata: issue command
net from https://kaplandm.github.io/stata
and follow instructions for installation (and email me if you have problems). Note: older Stata versions do not support https, in which case you can download the "code/tex/etc." .zip file above.
Note: for Stata version 11, you may need to use the sivqr11 command in file sivqr.ado

Frequentist properties of Bayesian inequality tests

2021, Journal of Econometrics
(with Longhao Zhuo)

published | accepted | code/tex/etc. | .bib

We characterize Bayesian and frequentist differences on general inequality hypotheses, even when credible/confidence sets coincide.

distcomp: Comparing distributions

2019, Stata Journal

published | accepted | code/tex/etc. | .bib

In Stata: issue command
net from https://kaplandm.github.io/stata
and follow instructions for installation (and email me if you have problems). Note: older Stata versions do not support https, in which case you can download the "code/tex/etc." .zip file above.

Smoothed GMM for quantile models

2019, Journal of Econometrics
(with Luciano de Castro, Antonio Galvao, and Xin Liu)

published | accepted | code/tex/etc. | .bib

We extend smoothed IVQR estimation (Kaplan and Sun, 2017) to non-iid data, nonlinear and over-identified models, with a quantile Euler equation empirical example.

Comparing distributions by multiple testing across quantiles or CDF values

2018, Journal of Econometrics
(with Matt Goldman)

published | Stata code/article | accepted | supplement | code | replication | .bib

Where do two distributions differ? We provide a new method to strongly control FWER while distributing power more evenly than Kolmogorov–Smirnov. One-sample and two-sample; stepdown and pre-test refinements; extension to conditional distributions and regression discontinuity.

Nonparametric inference on (conditional) quantile differences and linear combinations, using L-statistics

2018, The Econometrics Journal
(with Matt Goldman)
2018 Denis Sargan Econometrics Prize
Virtual issue on The Econometrics of Treatment Effects

published | accepted | supplement | replication | code:unconditional | code:conditional | .bib

We provide nonparametric, high-order accurate CIs for: quantile differences between two populations (which are QTEs under certain assumptions); interquantile ranges; more general linear combinations of quantiles (and differences thereof); and conditional (on X) versions of each.

Fractional order statistic approximation for nonparametric conditional quantile inference

2017, Journal of Econometrics
(with Matt Goldman)

published | accepted | supplement | code:unconditional | code:conditional | simulations | more sims | examples | .bib

We provide nonparametric CIs for quantiles and conditional quantiles, with high-order accuracy.

Smoothed estimating equations for instrumental variables quantile regression

2017, Econometric Theory
(with Yixiao Sun)

Adapted in Stata 18 as ivqregress smooth

published | accepted | R:estimator | replication | .bib

For IV quantile regression, we show smoothing improves computation and high-order properties.
See also sivqr Stata command/paper

Improved quantile inference via fixed-smoothing asymptotics and Edgeworth expansion

2015, Journal of Econometrics

published | accepted | appendix 1 | appendix 2 | simulations | empirical | R code | R examples | MATLAB code | MATLAB examples | .bib

I study the Studentized sample quantile: fixed-smoothing asymptotics is more accurate and suggests an "inference-optimal" bandwidth to maximize accuracy; practical advantage biggest near tails.

A computational approach to style in American poetry

2007, International Conference on Data Mining (ICDM)
(with David Blei)

published | accepted | longer draft | slides | code/app | .bib

I analyze poetic texts by extracting orthographic, syntactic, and phonemic features, visualizing and comparing poems in the corresponding vector space of features. ("One of the most thorough and sophisticated computing analysis of poems to date" rave Wang and Yang, 2015.)

Resting and Superseded Projects

2021, Relationship Estimators
paper | code/tex/etc. | .bib
Highlighting the difference between "my estimated parameter satisfies the relationship" and "I estimate that the relationship holds."

2020, Assessing Policy Effects with Unconditional Quantile Regression
I (eventually) noticed this was essentially the same as Proposition 1 of Rothe (2010).

2019, Unbiased Estimation as a Public Good
paper | code/tex/etc. | .bib
Under certain conditions, an estimator's bias is relatively more important (compared to its variance) when contributing to the public body of scientific knowledge than for a single estimate.
(I received many helpful ideas and references from referees that I enjoyed reading and considering, but I'm not sure if/when I'll get back to this project.)

2019, Optimal Smoothing in Divide-and-Conquer for Big Data
My main point is not new; for example, see "Related work" on the second page of Xu, Shang, and Cheng (2018), "Optimal Tuning for Divide-and-conquer Kernel Ridge Regression with Massive Data".

2015, Bayesian and frequentist tests of sign equality and other nonlinear inequalities
paper | .bib
Compares various tests of "sign equality" (hypothesis that two parameters have the same +/- sign), including a new (but strange) nearly-unbiased test, as well as LR, Wald, and Bayes. The analysis of more general hypothesis testing of regions was later published the Journal of Econometrics paper "Frequentist properties of Bayesian inequality tests" above.

2014, Nonparametric inference on quantile marginal effects
paper | code | simulations | example | .bib
It works well enough, but honestly I'd just bootstrap it instead of using my method.

2013, IDEAL inference on conditional quantiles: superseded by above papers "Fractional order statistic approximation for nonparametric conditional quantile inference" and "Nonparametric inference on conditional quantile treatment effects using L-statistics"

2013, IDEAL quantile inference via interpolated duals of exact analytic L-statistics: superseded by above papers "Fractional order statistic approximation for nonparametric conditional quantile inference" and "Nonparametric inference on conditional quantile treatment effects using L-statistics"

2011, Fixed-smoothing asymptotics and accurate F approximation using vector autoregressive variance matrix estimator (with Yixiao Sun)
paper | .bib

Experiencing technical difficulties (error in proofs).

For curious grad students:
work I did in grad school

2009, summer research report examining data-dependent methods for sieve size selection in nonparametric IV estimation.

2010, Natural disasters and differential household effects: evidence from the May 2006 Java earthquake
paper | slides | .bib

Were poorer households hurt more? Examining direct and indirect mechanisms.

2013, dissertation and dissertation defense