Pith. sign in

REVIEW 2 major objections 6 minor 72 references

Inferring Effects of Major Events through Discontinuity Forecasting of Population Anxiety

T0 review · 2 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a community's anxiety response to a major event can be predicted before the event, from pre-event mental-health trajectories and static community text embeddings.

desk verdict New forecasting task, real predictive signal, but the 'exogenous' county embeddings leak the outcome, so the headline improvement is suspect. read the letter →

arxiv 2508.21722 v1 pith:4QUCIDJB submitted 2025-08-29 cs.LG

Siddharth Mangalik , Ojas Deshpande , Adithya V. Ganesan , Sean A. P. Clouston , H. Andrew Schwartz This is my paper · ORCID
classification cs.LG
keywords discontinuityforecastingregressiondesignlanguage-basedmentalhealthassessmentsanxietyCOVID-19county-levelpredictioncausalinferencefromtexttime-series
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the mental-health effect of a major event on a specific community is not just measurable after the fact but forecastable in advance. It reframes the econometric Longitudinal Regression Discontinuity Design as a supervised learning task: for each US county, the size of the anxiety jump (discontinuity) and the change in post-event anxiety trend (slope) at the arrival of COVID-19 are computed from before/after linear fits, then predicted from data available before the event. Best held-out correlations reach roughly r=.76 for the jump and r=.87 for the slope, well above a no-change baseline, and the paper reports that adding county-level language embeddings and depression trajectories improves the strongest models. If right, public-health agencies could anticipate which communities will respond most intensely to a pandemic or comparable shock and pre-position support.

What carries the argument

Longitudinal Regression Discontinuity Design (LRDD) deltas: for each county, linear regressions fit to a 9-week pre-event window and a 9-week post-event window produce an intercept difference delta_0 (the discontinuity, i.e., the event-time shift in anxiety) and a slope difference delta_1 (the change in weekly trend). These two numbers become the regression targets. The model inputs are the pre-event anxiety series, its fitted before-period beta_0 and beta_1, dynamic covariates (depression timeline and its coefficients), and static county embeddings obtained by mean-aggregating pretrained language-model tweet representations. The LRDD step is what converts an observational time series into a

What would settle it

Retrain the best model with county embeddings computed only from tweets dated before the earliest event in the study (e.g., all of 2019 only) and compare held-out correlations. If the exogenous-embedding gain largely disappears, the reported predictive power is not fully prospective. A second check: run the same pipeline on a placebo event with no expected effect; if held-out r remains high for a null-event delta, the evaluation protocol is detecting label noise rather than genuine discontinuities.

Watch

Extended reading notes

Core claim

The central claim is that future discontinuities are learnable: (delta_0, delta_1) = f(pre-event anxiety series, its fitted linear coefficients, dynamic covariates like depression, and static exogenous county embeddings) is a well-posed multi-output regression, and it works out-of-sample for the first US county COVID-19 cases. The paper evaluates this on 361 counties using weekly language-based anxiety assessments, finds the task difficult but increasingly feasible as model sophistication grows, and reports best held-out Pearson correlations of about .76 for the discontinuity and .87 for the slope. A placebo test with randomized event dates yields mean discontinuity near zero, supporting the

Load-bearing premise

The county embeddings called 'exogenous' are built from tweets spanning 2019-2020, the same period in which weekly anxiety is measured, so the claim that they represent pre-event community context assumes they contain no post-event or outcome-related linguistic signal.

Editorial extensions

If this is right

  • Held-out correlations of about .76 for discontinuity and .87 for slope mean a county's idiosyncratic anxiety response to a pandemic arrival is learnable before the event happens.
  • The same discontinuity-forecasting setup could be retrained for other event types, such as natural disasters, mass violence, or policy changes, as long as dense language-based assessments are available for training; the paper explicitly warns that the current model does not generalize across event types.
  • Since the control-event randomization yields near-zero mean deltas, the LRDD labels isolate event-related shifts rather than background trends, giving the prediction task a causal target.
  • The finding that high-SES and high-urbanicity counties show larger average spikes but less predictable responses sets a bound on where such forecasts can be used with confidence.
  • Discontinuity forecasting changes the policy use of mental-health surveillance: instead of detecting a spike after it happens, agencies could allocate resources before an expected event.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A strict test of the exogenous embeddings is the paper's own cited best practice of using disjoint text for causal measurement; building county embeddings from a pre-event-only corpus, such as 2019 tweets only, would tell whether the reported +exog gain is prospective or reflects leakage from post-event or outcome-related language. The paper does not run that test.
  • The same delta_0 and delta_1 targets could be computed for depression, which the paper already includes as a dynamic covariate, so a joint multi-outcome discontinuity-forecasting model for anxiety and depression is a natural extension.
  • If linear before/after fits are a poor approximation for counties with nonlinear pre-trends, the delta labels become noisy; using locally linear or kernel-weighted RDD estimates would test whether the forecastable signal is an artifact of the linear specification.
  • The method's promise for hypothetical events depends on the stability of community text embeddings over time; an event that changes a community's composition or digital behavior could break the mapping learned from 2019-2020 data.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper introduces 'discontinuity forecasting': a supervised learning task in which the size of a community's regression-discontinuity jump (δ0) and slope change (δ1) in language-based anxiety after a major event (the first COVID-19 case in a US county) is predicted from pre-event data. The inputs are the pre-event anxiety series, its fitted regression coefficients, depression timelines as dynamic covariates, and 'exogenous' RoBERTa-based county embeddings. The authors compare GOFAI, ensemble, GRU, and Transformer models, reporting held-out Pearson correlations up to .76 for δ0 and .87 for δ1, and they validate their LRDD implementation with randomized control events that yield near-zero mean discontinuities. They also analyze how performance varies by SES and urbanicity.

Significance. The proposed task is novel and could be practically valuable for anticipating community-specific mental-health responses to public-health events. The paper has several strengths: a sensible suite of baselines (no-change, mean, ARIMA forecasting), a randomized-event control that supports the LRDD measurement, and consistent performance patterns across model families. However, the central claim that effects can be predicted 'from pre-event data alone' is compromised by the construction of the exogenous county embeddings and by the normalization of the outcome and input series. These leakage paths mean that the headline improvements from including 'exogenous' variables are not currently supported. If the authors can reconstruct the exogenous inputs from pre-event text only and re-estimate normalization statistics without post-event information, the contribution would be significant.

major comments (2)
  1. [§4.3, Eq. (1), Tables 2–3] The 'exogenous' county embeddings are constructed by mean-aggregating RoBERTa representations of every tweet in the County Tweet Lexical Bank from 2019 to 2020. This is the same corpus and time window from which the LBMHA anxiety outcome (Section 4.1) is measured. Since each county's first COVID-19 case occurs within 2020, the embeddings for essentially every county include tweets written after the event date, and these tweets contain anxiety-related language that is precisely the signal used to estimate δ0 and δ1. The problem formulation in Section 3 explicitly requires 'no post-event data' in the inputs, but this condition is violated. Consequently, the performance gains attributed to '+exog' (e.g., Transformer r=.678→.753; GRU r=.713→.755, Tables 2–3) may reflect direct target leakage rather than the value of external context. The manuscript itself cites Egami et al. [54] and Feder et
  2. [§4.1] The per-county Z-score normalization uses the full 2019–2020 distribution: scores are 'county mean-centered at 0' with a standard deviation estimated over the entire period. Because the same period contains the post-event window, the scaled pre-event inputs y_{−T:t−1} carry information about post-event variability. The mean cancels in the definition of δ0, but the standard deviation does not; it scales both the regressors and the slope estimates. This is an additional leakage path for the 'pre-event data alone' claim. The authors should estimate normalization statistics from pre-event data only, or use raw scores, and then re-run the analyses.
minor comments (6)
  1. [Throughout] There are numerous typos and grammatical slips, e.g., 'resaerchers', 'poo part', 'capabale', 'futher', 'a examination', and 'diff. in diff.' Some of these impede readability and should be corrected.
  2. [§3] The notation y_{−T:t−1} is not defined clearly; it should explicitly denote the sequence y_{−T}, y_{−T+1}, ..., y_{t−1}. Similarly, 'endogenous timeline variables' is ambiguous; specify what is included.
  3. [Abstract and §4.3] The claim 'strong improvement (r=+.46 for discontinuity and r=+.65 for slope) over traditional static community representations' is not sufficiently precise. Which baseline is this compared to, and where are the corresponding numbers in the tables? Table 5 suggests that exog features alone perform poorly (r=.257 for δ0), so the basis for this improvement claim should be stated explicitly.
  4. [Tables 2 and 3] Significance asterisks are sparse and it is unclear what comparison is being tested (e.g., base vs '+exog', base vs '+exog+cov'). State the null hypothesis and the multiple-testing correction, if any.
  5. [Appendix D.1] The mean baseline description says 'the mean intercept (δ̄0) and slope (δ̄1) discontinuity of the training set'—presumably this is the mean of the training labels, not a fitted intercept. Clarify the wording.
  6. [§4.3 and Table 5] The term 'static representations' for county embeddings is misleading because these embeddings are aggregated from 2019–2020 text and are therefore time-indexed. This is related to the major leakage concern and should be clarified in the text.

Circularity Check

1 steps flagged · score 6.0 of 10

Exogenous county embeddings are computed from the same 2019–2020 tweet corpus as the anxiety outcome, so the '+exog' gains may reflect post-event target leakage rather than external context.

  1. self definitional [Section 3 (Problem Formulation), Section 4.1 (Language-Based Mental Health Assessments), Section 4.3 (Predictive Models)]
    "given a history of the endogenous outcome preceding the event ( y−T :t−1), ... and no post-event data: (ˆδ0, ˆδ1) = f (y−T :t−1, β0,−T :t−1, β1,−T :t−1, Xexog, Xcov,−T :t−1) ... This dataset derives these assessments from ∼1 billion English tweets made in the US from 2019 to 2020 ... we create 1,422 county embeddings by first extracting the second-to-last layer RoBERTa-Largeembedding against every tweet in the County Tweet Lexical Bank [43, 22] from 2019 to 2020."

    The task is defined as using no post-event data, with Xexog described as 'static exogenous descriptors.' But the county embeddings used as Xexog are mean RoBERTa representations of every tweet from 2019 to 2020 — the same corpus and time window from which the LBMHA anxiety outcome is derived, and a window that extends past each county's first COVID-19 case in 2020. Post-event tweets containing anxiety-related language are therefore baked into the 'exogenous' input. Since the outcome δ0 and δ1 measure post-event anxiety shifts, the embedding can encode the target itself. The paper even cites Egami et al. [54] and Feder et al. [50] recommending disjoint text sets for causal measurement, but does not construct the embeddings from a pre-event-only corpus. Thus the reported improvements from '+

full rationale

The only significant circularity is the construction of the 'exogenous' county embeddings from the same 2019–2020 tweet stream that defines the LBMHA anxiety outcome, with the event occurring inside that window. This violates the paper's own no-post-event-data condition and creates target leakage; therefore the exogenous-contribution claims are partially circular. The endogenous-only forecasting chain (pre-event anxiety history and regression coefficients to predict δ0/δ1) is not circular: those inputs are genuinely pre-event and the baselines are reasonable. The self-citation to LBMHA [22] is not load-bearing because the dataset is externally validated against Gallup and is an independent measurement source. The control-event validation is a legitimate non-circular check. Overall, the central 'no post-event data' guarantee is breached for the best-performing feature set, so the paper cannot claim that the +exog improvements reflect only external community context. Score 6 (partial circularity); not 8 because endogenous-only results remain meaningful and are not forced by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on several definitional choices (window length, buffer, inclusion threshold) and on three untested assumptions: the causal validity of the RDD target, the validity of the language-based outcome, and, most importantly, the exogeneity of county embeddings computed from the same temporal window as the outcome. The last assumption is ad hoc because the paper does not build a pre-event-only representation.

free parameters (5)
  • Event window length T = 9 weeks
    The definition of delta_0 and delta_1 in Section 3 uses 2T+1 = 19 weeks centered at the event; the choice of T=9 is made by hand and affects the target values.
  • Pre/post buffer = 1 week
    Section 4.2 applies a one-week buffer around the event to form pre- and post-envelope periods; Appendix B.1 shows sensitivity to 0, 1, and 2 weeks.
  • Reliability threshold for county-week inclusion = at least 200 unique users
    Section 4.1 filters county-week measurements to at least 200 unique individuals, the documented threshold from the LBMHA dataset; this restricts the analysis to 361 counties.
  • Anxiety Z-score normalization = county mean 0, sd 1
    Section 4.1 Z-scores per county, which changes the scale of the discontinuity values; the reported effect sizes are in within-county standard deviation units.
  • Model hyperparameters = see Appendix C Table 6
    All ML models have hyperparameters tuned on a dev set; these are standard fitting choices but are part of the pipeline that produces the reported r values.
assumptions (5)
  • domain assumption LRDD identification: anxiety would have continued smoothly around the event absent the shock, so discontinuities measure the event's effect.
    Section 3 and 4.4: the causal interpretation of delta_0 and delta_1 requires the smooth-counterfactual assumption; the random-control event only checks that random dates do not produce a discontinuity, not that the counterfactual holds for COVID.
  • domain assumption LBMHA weekly anxiety scores are valid measures of population anxiety at county-week resolution.
    Section 4.1: the entire analysis uses the authors' previous LBMHA dataset [22]; validity rests on convergent validity with surveys and not on any independent clinical measure.
  • ad hoc to paper The 2019-2020 county text embeddings are exogenous to the outcome and do not contain post-event information.
    Section 4.3: the embeddings are mean-aggregated RoBERTa representations of tweets from 2019-2020, the same window as the anxiety outcome; no time-disjoint validation is provided despite the paper citing disjoint-text best practices.
  • domain assumption Twitter users can be reweighted to represent county populations.
    Section 4.1: the LBMHA weighting scheme is inherited from [22]; the representativeness of the weighted Twitter panel is not re-tested here.
  • standard math Standard regression and statistical learning techniques are valid for the estimation.
    Linear regressions for delta_0 and delta_1 and supervised models are standard; no new mathematics is introduced.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inferring Effects of Major Events through Discontinuity Forecasting of Population Anxiety." pith.science (2026). https://pith.science/paper/4QUCIDJB

@misc{pith2026250821722,
  author       = {Pith},
  title        = {Pith review of: Inferring Effects of Major Events through Discontinuity Forecasting of Population Anxiety},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4QUCIDJB}},
  note         = {Machine review of arXiv:2508.21722}
}
abstract

Estimating community-specific mental health effects of local events is vital for public health policy. While forecasting mental health scores alone offers limited insights into the impact of events on community well-being, quasi-experimental designs like the Longitudinal Regression Discontinuity Design (LRDD) from econometrics help researchers derive more effects that are more likely to be causal from observational data. LRDDs aim to extrapolate the size of changes in an outcome (e.g. a discontinuity in running scores for anxiety) due to a time-specific event. Here, we propose adapting LRDDs beyond traditional forecasting into a statistical learning framework whereby future discontinuities (i.e. time-specific shifts) and changes in slope (i.e. linear trajectories) are estimated given a location's history of the score, dynamic covariates (other running assessments), and exogenous variables (static representations). Applying our framework to predict discontinuities in the anxiety of US counties from COVID-19 events, we found the task was difficult but more achievable as the sophistication of models was increased, with the best results coming from integrating exogenous and dynamic covariates. Our approach shows strong improvement ($r=+.46$ for discontinuity and $r = +.65$ for slope) over traditional static community representations. Discontinuity forecasting raises new possibilities for estimating the idiosyncratic effects of potential future or hypothetical events on specific communities.

Figures

Figures reproduced from arXiv: 2508.21722 by the authors.

Figure 1
Figure 1. Calculation of discontinuity (δ0, green) and discontinuity slope (δ1, purple) before and after a critical event at t = 0, using regression parameters for the community assessment before (red) and after (teal) the event. 2 Contributions The main contributions of this work include: (1) A proposed regression discontinuity method to assess the effect of community-specific events on health assessments. (2) The evaluation… view at source ↗
Figure 2
Figure 2. Mean anxiety (a) and depression (b) before and after the first case of COVID-19 in US [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Individual lines show unique county anxiety assessment trends. To the right in green is the [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Anxiety before and after the first case of COVID-19 with no buffer (a) and a buffer size of [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Non-linear trends in Anxiety throughout 2019-2020. Results are stratified by counties into [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Discontinuities in Anxiety by Socio-Economic Status (a) and Urbanicity (b) before and [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Example of the diff-in-diff pattern of quasi-experimental design. Here we see the example [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 70 canonical work pages

  1. [54]

    Fong, Justin Grimmer, Margaret E

    Naoki Egami, Christian J. Fong, Justin Grimmer, Margaret E. Roberts, and Brandon M. Stewart. How to make causal inferences using texts. Science Advances, 8(42):eabg2652, 2022

    work page 2022

  2. [1]

    Dashboard, 2020

    WHO. Dashboard, 2020. Accessed 2020

    work page 2020

  3. [2]

    Characteristics of and public health responses to the coronavirus disease 2019 outbreak in china

    Sheng-Qun Deng and Hong-Juan Peng. Characteristics of and public health responses to the coronavirus disease 2019 outbreak in china. Journal of clinical medicine, 9(2):575, 2020

    work page 2019

  4. [3]

    Key lessons from the covid-19 public health response in australia

    JM Basseal, CM Bennett, Peter Collignon, BJ Currie, David N Durrheim, Julie Leask, ES McBryde, P McIntyre, FM Russell, DW Smith, et al. Key lessons from the covid-19 public health response in australia. The Lancet Regional Health–Western Pacific, 30, 2023

    work page 2023

  5. [4]

    The uk’s public health response to covid-19, 2020

    Gabriel Scally, Bobbie Jacobson, and Kamran Abbasi. The uk’s public health response to covid-19, 2020

    work page 2020

  6. [5]

    Global prevalence and burden of depressive and anxiety disorders in 204 coun- tries and territories in 2020 due to the covid-19 pandemic

    Damian F Santomauro, Ana M Mantilla Herrera, Jamileh Shadid, Peng Zheng, Charlie Ash- baugh, David M Pigott, Cristiana Abbafati, Christopher Adolph, Joanne O Amlag, Aleksandr Y Aravkin, et al. Global prevalence and burden of depressive and anxiety disorders in 204 coun- tries and territories in 2020 due to the covid-19 pandemic. The Lancet, 398(10312):170...

    work page 2020

  7. [6]

    Addressing the mental health impact of covid-19 through population health.Clinical psychology review, 85:102006, 2021

    Matt Boden, Lindsey Zimmerman, Kathryn J Azevedo, Josef I Ruzek, Sasha Gala, Hoda S Abdel Magid, Nichole Cohen, Robyn Walser, Naina D Mahtani, Katherine J Hoggatt, et al. Addressing the mental health impact of covid-19 through population health.Clinical psychology review, 85:102006, 2021

    work page 2021

  8. [7]

    Psychological outcomes associated with stay-at-home orders and the perceived impact of covid-19 on daily life

    Matthew T Tull, Keith A Edmonds, Kayla M Scamaldo, Julia R Richmond, Jason P Rose, and Kim L Gratz. Psychological outcomes associated with stay-at-home orders and the perceived impact of covid-19 on daily life. Psychiatry research, 289:113098, 2020

    work page 2020

Show all 72 references
  1. [8]

    Epidemiology of mental health problems in covid-19: a review

    Md Mahbub Hossain, Samia Tasnim, Abida Sultana, Farah Faizah, Hoimonty Mazumder, Liye Zou, E Lisako J McKyer, Helal Uddin Ahmed, and Ping Ma. Epidemiology of mental health problems in covid-19: a review. F1000Research, 9:636, 2020

    2020

  2. [9]

    Covid 19 and its mental health consequences, 2021

    Anant Kumar and K Rajasekharan Nayar. Covid 19 and its mental health consequences, 2021. 10

    2021

  3. [10]

    The impact of school closures on learning and mental health of children: Lessons from the covid-19 pandemic

    Deni Mazrekaj and Kristof De Witte. The impact of school closures on learning and mental health of children: Lessons from the covid-19 pandemic. Perspectives on Psychological Science, 19(4):686–693, 2024

    2024

  4. [11]

    School closures during social lockdown and mental health, health behaviors, and well-being among children and adolescents during the first covid-19 wave: a systematic review

    Russell Viner, Simon Russell, Rosella Saulle, Helen Croker, Claire Stansfield, Jessica Packer, Dasha Nicholls, Anne-Lise Goddings, Chris Bonell, Lee Hudson, et al. School closures during social lockdown and mental health, health behaviors, and well-being among children and ado...

    2022

  5. [12]

    Mental distress among us adults during the covid-19 pandemic

    Jean M Twenge and Thomas E Joiner. Mental distress among us adults during the covid-19 pandemic. Journal of clinical psychology, 76(12):2170–2182, 2020

    2020

  6. [13]

    Closure of universities due to coronavirus disease 2019 (covid-19): impact on education and mental health of students and academic staff

    Pradeep Sahu. Closure of universities due to coronavirus disease 2019 (covid-19): impact on education and mental health of students and academic staff. Cureus, 12(4), 2020

    2019

  7. [14]

    Mental health outcomes of the covid-19 pandemic

    Dalila Talevi, Valentina Socci, Margherita Carai, Giulia Carnaghi, Serena Faleri, Edoardo Trebbi, Arianna Di Bernardo, Francesco Capelli, and Francesca Pacitti. Mental health outcomes of the covid-19 pandemic. Rivista di psichiatria, 55(3):137–144, 2020

    2020

  8. [15]

    Impact of covid-19 pandemic on mental health in the general population: A systematic review

    Jiaqi Xiong, Orly Lipsitz, Flora Nasri, Leanna MW Lui, Hartej Gill, Lee Phan, David Chen-Li, Michelle Iacobucci, Roger Ho, Amna Majeed, et al. Impact of covid-19 pandemic on mental health in the general population: A systematic review. Journal of affective disorders, 277:55–64, 2020

    2020

  9. [16]

    Covid-19 pandemic and mental health conse- quences: Systematic review of the current evidence.Brain, behavior, and immunity, 89:531–542, 2020

    Nina Vindegaard and Michael Eriksen Benros. Covid-19 pandemic and mental health conse- quences: Systematic review of the current evidence.Brain, behavior, and immunity, 89:531–542, 2020

    2020

  10. [17]

    Sick individuals and sick populations

    Geoffrey Rose. Sick individuals and sick populations. International journal of epidemiology, 30(3):427–432, 2001

    2001

  11. [18]

    Public health surveillance: A tool for targeting and monitoring interventions

    Peter Nsubuga, Michael E White, Stephen B Thacker, et al. Public health surveillance: A tool for targeting and monitoring interventions. In Dean T Jamison, Joel G Breman, Anthony R Measham, et al., editors, Disease Control Priorities in Developing Countries, chapter 53. The In...

    2006

  12. [19]

    Loneliness across time and space

    Maike Luhmann, Susanne Buecker, and Marilena Rüsberg. Loneliness across time and space. Nature Reviews Psychology, 2(1):9–23, 2023

    2023

  13. [20]

    Epidemiologic evidence for the relation between socioeconomic status and depression, obesity, and diabetes

    Susan A Everson, Siobhan C Maty, John W Lynch, and George A Kaplan. Epidemiologic evidence for the relation between socioeconomic status and depression, obesity, and diabetes. Journal of psychosomatic research, 53(4):891–895, 2002

    2002

  14. [21]

    Socioeconomic inequalities in the spread of coronavirus-19 in the united states: A examination of the emergence of social inequalities

    Sean AP Clouston, Ginny Natale, and Bruce G Link. Socioeconomic inequalities in the spread of coronavirus-19 in the united states: A examination of the emergence of social inequalities. Social science & medicine, 268:113554, 2021

    2021

  15. [22]

    Ganesan, Shashanka Subrahmanya, Nikita Soni, Sean AP Clouston, et al

    Siddharth Mangalik, Johannes C Eichstaedt, Salvatore Giorgi, Jihu Mun, Farhan Ahmed, Gilvir Gill, Adithya V . Ganesan, Shashanka Subrahmanya, Nikita Soni, Sean AP Clouston, et al. Robust language-based mental health assessments in time and space through social media. NPJ Digit...

    2024

  16. [23]

    Regression-discontinuity analysis: An alternative to the ex post facto experiment

    Donald L Thistlethwaite and Donald T Campbell. Regression-discontinuity analysis: An alternative to the ex post facto experiment. Journal of Educational psychology, 51(6):309, 1960

    1960

  17. [24]

    Identification and estimation of treatment effects with a regression-discontinuity design

    Jinyong Hahn, Petra Todd, and Wilbert Van der Klaauw. Identification and estimation of treatment effects with a regression-discontinuity design. Econometrica, 69(1):201–209, 2001

    2001

  18. [25]

    The regression discon- tinuity design showed to be a valid alternative to a randomized controlled trial for estimating treatment effects

    Iris L Maas, Sandra Nolte, Otto B Walter, Thomas Berger, Martin Hautzinger, Fritz Hohagen, Wolfgang Lutz, Björn Meyer, Johanna Schröder, Christina Späth, et al. The regression discon- tinuity design showed to be a valid alternative to a randomized controlled trial for estimati...

    2017

  19. [26]

    Timexer: Empowering transformers for time series forecasting with exogenous variables

    Yuxuan Wang, Haixu Wu, Jiaxiang Dong, Guo Qin, Haoran Zhang, Yong Liu, Yunzhong Qiu, Jianmin Wang, and Mingsheng Long. Timexer: Empowering transformers for time series forecasting with exogenous variables. arXiv preprint arXiv:2402.19072, 2024

    2024 arXiv

  20. [27]

    Ada-mshyper: adaptive multi- scale hypergraph transformer for time series forecasting

    Zongjiang Shang, Ling Chen, Binqing Wu, and Dongliang Cui. Ada-mshyper: adaptive multi- scale hypergraph transformer for time series forecasting. Advances in Neural Information Processing Systems, 37:33310–33337, 2024

    2024

  21. [28]

    Let’s take the con out of econometrics

    Edward E Leamer. Let’s take the con out of econometrics. The American Economic Review, 73(1):31–43, 1983

    1983

  22. [29]

    The use and interpretation of quasi-experimental studies in medical informatics

    Anthony D Harris, Jessina C McGregor, Eli N Perencevich, Jon P Furuno, Jingkun Zhu, Dan E Peterson, and Joseph Finkelstein. The use and interpretation of quasi-experimental studies in medical informatics. Journal of the American Medical Informatics Association, 13(1):16–23, 2006

    2006

  23. [30]

    Mostly harmless econometrics: An empiricist’s companion

    Joshua D Angrist and Jörn-Steffen Pischke. Mostly harmless econometrics: An empiricist’s companion. Princeton university press, 2009

    2009

  24. [31]

    Quantifying causality in data science with quasi- experiments

    Tony Liu, Lyle Ungar, and Konrad Kording. Quantifying causality in data science with quasi- experiments. Nature computational science, 1(1):24–32, 2021

    2021

  25. [32]

    Ensuring causal, not casual, inference

    Rashelle J Musci and Elizabeth Stuart. Ensuring causal, not casual, inference. Prevention Science, 20:452–456, 2019

    2019

  26. [33]

    Regression discontinuity designs: A guide to practice

    Guido W Imbens and Thomas Lemieux. Regression discontinuity designs: A guide to practice. Journal of econometrics, 142(2):615–635, 2008

    2008

  27. [34]

    Regression discontinuity designs in economics

    David S Lee and Thomas Lemieux. Regression discontinuity designs in economics. Journal of economic literature, 48(2):281–355, 2010

    2010

  28. [35]

    Covid-19 panel microdata, 2021

    Gallup. Covid-19 panel microdata, 2021

    2021

  29. [36]

    Coronavirus (covid-19) data in the united states

    NYT. Coronavirus (covid-19) data in the united states. https://github.com/nytimes/ covid-19-data, 2023

    2023

  30. [37]

    Comorbid depression and anxiety spectrum disorders

    Jack M Gorman. Comorbid depression and anxiety spectrum disorders. Depression and anxiety, 4(4):160–168, 1996

    1996

  31. [38]

    Comorbid anxiety and depression

    Mark H Pollack. Comorbid anxiety and depression. Journal of Clinical Psychiatry, 66:22, 2005

    2005

  32. [39]

    Treating comorbid anxiety and depression: Psychosocial and pharmacological approaches

    Jeremy D Coplan, Cindy J Aaronson, Venkatesh Panthangi, and Younsuk Kim. Treating comorbid anxiety and depression: Psychosocial and pharmacological approaches. World journal of psychiatry, 5(4):366, 2015

    2015

  33. [40]

    For high-dimensional hierarchical mod- els, consider exchangeability of effects across covariates instead of across datasets

    Brian Trippe, Hilary Finucane, and Tamara Broderick. For high-dimensional hierarchical mod- els, consider exchangeability of effects across covariates instead of across datasets. Advances in Neural Information Processing Systems, 34:13471–13484, 2021

    2021

  34. [41]

    Predicting economic growth by region embedding: A multigraph convolutional network approach

    Bo Hui, Da Yan, Wei-Shinn Ku, and Wenlu Wang. Predicting economic growth by region embedding: A multigraph convolutional network approach. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, pages 555–564, 2020

    2020

  35. [42]

    Census2vec: Enhancing socioeconomic predictive models with geo-embedded data

    Ravi Varma Kumar Bevara, Isabelle Wagenvoord, Farahnaz Hosseini, Himanshu Sharma, Vandana Nunna, and Ting Xiao. Census2vec: Enhancing socioeconomic predictive models with geo-embedded data. In Intelligent Systems Conference, pages 626–640. Springer, 2024

    2024

  36. [43]

    Ungar, and H

    Salvatore Giorgi, Daniel Preotiuc-Pietro, Anneke Buffone, Daniel Rieman, Lyle H. Ungar, and H. Andrew Schwartz. The remarkable benefit of user-level aggregation for lexical-based population-level predictions. In Proceedings of the 2018 Conference on Empirical Methods in Natura...

    2018

  37. [44]

    Andrew Schwartz

    Salvatore Giorgi, Daniel Preo¸ tiuc-Pietro, Anneke Buffone, Daniel Rieman, Lyle Ungar, and H. Andrew Schwartz. The remarkable benefit of user-level aggregation for lexical-based population-level predictions. In Ellen Riloff, David Chiang, Julia Hockenmaier, and Jun’ichi Tsujii...

    2018

  38. [45]

    A theory of human motivation

    Abraham Harold Maslow. A theory of human motivation. Psychological review, 50(4):370, 1943

    1943

  39. [46]

    Exploring the relationship between money attitudes and maslow’s hierarchy of needs

    Mark Oleson. Exploring the relationship between money attitudes and maslow’s hierarchy of needs. International journal of consumer studies, 28(1):83–92, 2004

    2004

  40. [47]

    Causal inference and counterfactual prediction in machine learning for actionable healthcare

    Mattia Prosperi, Yi Guo, Matt Sperrin, James S Koopman, Jae S Min, Xing He, Shannan Rich, Mo Wang, Iain E Buchan, and Jiang Bian. Causal inference and counterfactual prediction in machine learning for actionable healthcare. Nature Machine Intelligence, 2(7):369–375, 2020

    2020

  41. [48]

    Explaining machine learning classifiers through diverse counterfactual explanations

    Ramaravind K Mothilal, Amit Sharma, and Chenhao Tan. Explaining machine learning classifiers through diverse counterfactual explanations. In Proceedings of the 2020 conference on fairness, accountability, and transparency, pages 607–617, 2020

    2020

  42. [49]

    Dickerson, and Keegan Hines

    Sahil Verma, John P. Dickerson, and Keegan Hines. Counterfactual explanations for machine learning: A review. CoRR, abs/2010.10596, 2020

    2010 arXiv

  43. [50]

    Causal inference in natural language processing: Estimation, prediction, interpretation and beyond

    Amir Feder, Katherine A Keith, Emaad Manzoor, Reid Pryzant, Dhanya Sridhar, Zach Wood- Doughty, Jacob Eisenstein, Justin Grimmer, Roi Reichart, Margaret E Roberts, et al. Causal inference in natural language processing: Estimation, prediction, interpretation and beyond. Transa...

    2022

  44. [51]

    Regression discontinuity designs with unknown discontinuity points: Testing and estimation

    Jack Porter and Ping Yu. Regression discontinuity designs with unknown discontinuity points: Testing and estimation. Journal of Econometrics, 189(1):132–147, 2015

    2015

  45. [52]

    Auto- mated local regression discontinuity design discovery

    William Herlands, Edward McFowland III, Andrew Gordon Wilson, and Daniel B Neill. Auto- mated local regression discontinuity design discovery. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pages 1512–1520, 2018

    2018

  46. [53]

    Automated detec- tion of causal inference opportunities: regression discontinuity subgroup discovery.Transactions on Machine Learning Research, 2023

    Tony Liu, Patrick Lawlor, Lyle Ungar, Konrad Kording, and Rahul Ladhania. Automated detec- tion of causal inference opportunities: regression discontinuity subgroup discovery.Transactions on Machine Learning Research, 2023

    2023

  47. [55]

    Forecasting treatment responses over time using recurrent marginal structural networks

    Bryan Lim. Forecasting treatment responses over time using recurrent marginal structural networks. Advances in neural information processing systems, 31, 2018

    2018

  48. [56]

    Estimating coun- terfactual treatment outcomes over time through adversarially balanced representations

    Ioana Bica, Ahmed M Alaa, James Jordon, and Mihaela van der Schaar. Estimating coun- terfactual treatment outcomes over time through adversarially balanced representations. In International Conference on Learning Representations, 2019

    2019

  49. [57]

    G-net: a deep learning approach to g-computation for counterfactual outcome prediction under dynamic treatment regimes

    Rui Li, Zach Shahn, Jun Li, Mingyu Lu, Prithwish Chakraborty, Daby Sow, Mohamed Ghalwash, and Li-wei H Lehman. G-net: a deep learning approach to g-computation for counterfactual outcome prediction under dynamic treatment regimes. arXiv preprint arXiv:2003.10551, 2020

    2003 arXiv

  50. [58]

    Causal transformer for estimating counterfactual outcomes

    Valentyn Melnychuk, Dennis Frauen, and Stefan Feuerriegel. Causal transformer for estimating counterfactual outcomes. In International conference on machine learning, pages 15293–15329. PMLR, 2022

    2022

  51. [59]

    Causal contrastive learning for counterfactual regression over time

    Mouad El Bouchattaoui, Myriam Tami, Benoit Lepetit, and Paul-Henry Cournède. Causal contrastive learning for counterfactual regression over time. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors, Advances in Neural Information Proce...

    2024

  52. [60]

    Counterfactual explanations and algorithmic recourses for machine learning: A review

    Sahil Verma, Varich Boonsanong, Minh Hoang, Keegan Hines, John Dickerson, and Chirag Shah. Counterfactual explanations and algorithmic recourses for machine learning: A review. ACM Comput. Surv., 56(12), October 2024

    2024

  53. [61]

    The emotional and mental health impact of the murder of george floyd on the us population.Proceedings of the National Academy of Sciences, 118(39):e2109139118, 2021

    Johannes C Eichstaedt, Garrick T Sherman, Salvatore Giorgi, Steven O Roberts, Megan E Reynolds, Lyle H Ungar, and Sharath Chandra Guntuku. The emotional and mental health impact of the murder of george floyd on the us population.Proceedings of the National Academy of Sciences,...

    2021

  54. [62]

    Geographic disparities in children’s mental health care

    Roland Sturm, Jeanne S Ringel, and Tatiana Andreyeva. Geographic disparities in children’s mental health care. Pediatrics, 112(4):e308–e308, 2003

    2003

  55. [63]

    Estimating geographic subjective well-being from twitter: A comparison of dictionary and data-driven language methods

    Kokil Jaidka, Salvatore Giorgi, H Andrew Schwartz, Margaret L Kern, Lyle H Ungar, and Johannes C Eichstaedt. Estimating geographic subjective well-being from twitter: A comparison of dictionary and data-driven language methods. Proceedings of the national academy of sciences, ...

    2020

  56. [64]

    Linguistic markers of psy- chological change surrounding september 11, 2001

    Michael A Cohn, Matthias R Mehl, and James W Pennebaker. Linguistic markers of psy- chological change surrounding september 11, 2001. Psychological science, 15(10):687–693, 2004

    2001

  57. [65]

    The ripple of fear, sympathy and solidarity during the boston bombings

    Yu-Ru Lin and Drew Margolin. The ripple of fear, sympathy and solidarity during the boston bombings. EPJ Data Science, 3:1–28, 2014

    2014

  58. [66]

    social desirability

    Derek L Phillips and Kevin J Clancy. Some effects of" social desirability" in survey studies. American journal of sociology, 77(5):921–940, 1972

    1972

  59. [67]

    Selection bias in web surveys

    Jelke Bethlehem. Selection bias in web surveys. International statistical review, 78(2):161–188, 2010

    2010

  60. [68]

    Changing sense of place in privately owned public spaces during the pandemic: Evidence from a regression discontinuity in time design

    Eunah Jung. Changing sense of place in privately owned public spaces during the pandemic: Evidence from a regression discontinuity in time design. Environmental Impact Assessment Review, 110:107706, 2025

    2025

  61. [69]

    The use and misuse of counterfactuals in ethical machine learning

    Atoosa Kasirzadeh and Andrew Smart. The use and misuse of counterfactuals in ethical machine learning. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, page 228–236, New York, NY , USA, 2021. Association for Computing Machinery

    2021

  62. [70]

    Burden of disease scenarios for 204 countries and territories, 2022–2050: a forecasting analysis for the global burden of disease study 2021

    Stein Emil V ollset, Hazim S Ababneh, Yohannes Habtegiorgis Abate, Cristiana Abbafati, Rouzbeh Abbasgholizadeh, Mohammadreza Abbasian, Hedayat Abbastabar, Abdallah HA Abd Al Magied, Samar Abd ElHafeez, Atef Abdelkader, et al. Burden of disease scenarios for 204 countries and t...

    2022

  63. [71]

    Economic burden of anxiety disorders: a systematic review and meta-analysis

    Alexander Konnopka and Hannah König. Economic burden of anxiety disorders: a systematic review and meta-analysis. Pharmacoeconomics, 38:25–37, 2020

    2020

  64. [72]

    Trends in population mental health before and after the 2008 recession: a repeat cross-sectional analysis of the 1991–2010 health surveys of england

    Srinivasa Vittal Katikireddi, Claire L Niedzwiedz, and Frank Popham. Trends in population mental health before and after the 2008 recession: a repeat cross-sectional analysis of the 1991–2010 health surveys of england. BMJ open, 2(5):e001790, 2012. 14 Figure 3: Individual line...

    2008

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.