pith. sign in

arxiv: 2605.19229 · v1 · pith:7V3XX7FTnew · submitted 2026-05-19 · 💻 cs.AI

Can Large Language Models Revolutionize Survey Research? Experiments with Disaster Preparedness Responses

Yan Wang , Ziyi Guo , Christopher McCarty This is my paper

Pith reviewed 2026-05-20 06:26 UTC · model grok-4.3

classification 💻 cs.AI
keywords large language modelssurvey researchmissing data imputationdisaster preparednessProtection Motivation Theoryblock-wise missingnessMNAR
0
0 comments X

The pith

A theory-anchored LLM reduces error and bias when imputing missing disaster survey responses better than standard statistical methods.

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

The paper evaluates large language models across the full survey workflow, focusing on their use for imputing block-wise missing data in a real Florida hurricane preparedness survey. It introduces a five-stage integration framework and tests several LLM setups against classical imputation techniques under realistic missing-not-at-random conditions. The central innovation anchors LLM retrieval in a knowledge graph derived from Protection Motivation Theory so that inferences respect established causal links among preparedness behaviors. This anchored approach delivered lower root-mean-square error and near-zero overall bias compared with inverse-probability weighting, multiple imputation, and random-forest methods. The authors also demonstrate that low aggregate bias can conceal opposing errors within subgroups and therefore recommend routine stratified bias checks.

Core claim

Organizing retrieval around PMT causal structure and integrating all evidence in a single model call outperforms unstructured retrieval and staged sequential inference. The proposed Anchored Marginal Theory-Informed LLM (A-TLM) outperforms all three classical imputation baselines (IPW/MI, MICE+PMM, missForest) on RMSE under disaster-relevant block-wise MNAR conditions (S4 RMSE 1.439 vs. 1.496 for the next-best), while achieving near-zero signed bias (-0.121) where the random-forest imputer produces the largest absolute bias (-0.631).

What carries the argument

The Anchored Marginal Theory-Informed LLM (A-TLM), which grounds LLM retrieval in a Protection Motivation Theory-constrained co-occurrence knowledge graph to produce imputations that respect documented causal relationships among survey items.

Load-bearing premise

A knowledge graph built from Protection Motivation Theory correctly represents the causal structure of how people answer disaster preparedness questions and does not introduce new systematic errors when used to guide the model.

What would settle it

Collect follow-up responses from the original respondents who had missing data and compare the distribution of A-TLM imputations against those actual answers to test whether the error and bias advantages hold outside the original sample.

Figures

Figures reproduced from arXiv: 2605.19229 by Christopher McCarty, Yan Wang, Ziyi Guo.

Figure 1
Figure 1. Figure 1: Theory-constrained staged prediction pipeline. The six￾stage PMT cascade maps Block A demographic inputs through tem￾porally ordered constructs to hurricane preparation outcomes. Edges retained in the PMT-constrained graph are shown; validated edges (Spearman sign-concordant) are highlighted. spanning routine time allocation, perceived time constraints, and preparation timing. It does not, however, provide… view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Bias-RMSE frontier on Block C imputation by missingness scenario. Methods below and to the left of A-TLM are preferable on both metrics simultaneously; no method clearly dominates A-TLM in any scenario. achieved the lowest RMSE in S1, S2, and S3 but the largest ab￾solute bias of any method in S4 (−0.631 ordinal levels), reflect￾ing a systematic under-prediction of compound-vulnerable respondents that would… view at source ↗
Figure 5
Figure 5. Figure 5: Captured exchanges from the Graph-Grounded Survey Assistant. Top: successful quantitative retrieval with PMT-stage contextualization. Bottom: epistemic refusal when retrieved evidence is insufficient. 6. Discussion 6.1 Contributions to Knowledge This paper operationalizes a five-stage framework for em￾bedding large language models within the survey-research workflow and evaluates each stage on the 2024 Hur… view at source ↗
read the original abstract

Survey research faces mounting structural challenges: declining response rates, sample bias, block-wise missingness among at-risk respondents, and AI-assisted fraudulent completions in online panels. Large language models (LLMs) have been proposed as a remedy, yet rigorous evaluations across the full survey workflow remain scarce, particularly in disaster contexts where data quality matters most. We present and evaluate a five-stage framework for LLM integration covering questionnaire design, sample selection, pilot testing, missing-data imputation, and post-collection analysis, using the 2024 Hurricane Milton preparedness survey of Florida residents (n=946) as a shared empirical testbed. We introduce a Protection Motivation Theory (PMT)-constrained co-occurrence knowledge graph and develop seven LLM configurations spanning zero-shot inference, retrieval-augmented baselines, and novel theory-informed variants. Our proposed Anchored Marginal Theory-Informed LLM (A-TLM) outperforms all three classical imputation baselines (IPW/MI, MICE+PMM, missForest) on RMSE under disaster-relevant block-wise MNAR conditions (S4 RMSE 1.439 vs. 1.496 for the next-best), while achieving near-zero signed bias (-0.121) where the random-forest imputer produces the largest absolute bias (-0.631). Organizing retrieval around PMT causal structure and integrating all evidence in a single model call outperforms unstructured retrieval and staged sequential inference (MAE 0.993 vs. 1.097 for standard RAG). We document that near-zero aggregate bias can mask opposing subgroup errors and propose subgroup-stratified bias auditing as a reporting standard. A retrieval-constrained knowledge-graph chatbot demonstrates that hallucination is architecturally manageable through grounded refusal.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The manuscript presents a five-stage framework for integrating large language models into survey research workflows, evaluated on the 2024 Hurricane Milton preparedness survey of Florida residents (n=946). It introduces a Protection Motivation Theory (PMT)-constrained co-occurrence knowledge graph and seven LLM configurations, with the proposed Anchored Marginal Theory-Informed LLM (A-TLM) claimed to outperform classical imputation baselines (IPW/MI, MICE+PMM, missForest) on RMSE (1.439 vs. 1.496 for the next-best) and signed bias (-0.121 vs. -0.631) under block-wise MNAR conditions, while also addressing hallucination via grounded refusal and proposing subgroup-stratified bias auditing.

Significance. If the empirical claims hold without artifacts from data leakage or insufficient controls, the work could meaningfully advance LLM applications in survey methodology by demonstrating theory-grounded retrieval augmentation for imputation in high-stakes disaster contexts. It provides concrete metrics from a real survey testbed, compares against established baselines, and highlights practical issues like masked subgroup errors. The approach of anchoring on PMT structure is a strength if the graph supplies independent causal information; otherwise the gains may not generalize beyond this setup.

major comments (3)
  1. [Methods (PMT-constrained co-occurrence knowledge graph and A-TLM description)] The construction of the PMT-constrained co-occurrence knowledge graph is not described in sufficient detail to determine its data sources or independence from the n=946 survey responses used for imputation testing. If edges or co-occurrences are extracted from the same response patterns (even under PMT constraints), retrieval-augmented generation could indirectly access the block-wise missingness patterns being imputed, unlike the classical baselines (MICE+PMM, missForest) that receive no such information. This directly threatens the central claim that A-TLM's RMSE advantage (S4: 1.439) and near-zero bias arise from theory-informed structure rather than leakage. Please specify the exact sources, construction process, and any overlap with the test data.
  2. [Experimental Setup and LLM Configurations] Exact prompt templates for the seven LLM configurations, the process for integrating the knowledge graph into retrieval, and whether data splits, exclusions, or MNAR simulation parameters were pre-specified versus post-hoc are not provided. Without these, the reported performance numbers (e.g., A-TLM RMSE 1.439 and bias -0.121 under block-wise MNAR) cannot be fully assessed for reproducibility or robustness against implementation choices.
  3. [Results (bias and subgroup analysis)] The manuscript notes that near-zero aggregate bias can mask opposing subgroup errors and proposes subgroup-stratified auditing as a reporting standard, but it is unclear whether this auditing was actually applied to the A-TLM results or only suggested. If not performed, the superiority claim over missForest (largest absolute bias -0.631) remains incomplete, as subgroup-specific errors could undermine the practical utility in disaster preparedness contexts.
minor comments (3)
  1. [Abstract] The abstract references 'S4 RMSE' without defining the scenario or table/figure it corresponds to; add a brief clarification or cross-reference.
  2. [Introduction and Methods] The acronym 'A-TLM' is introduced without an explicit expansion or definition of the 'Anchored Marginal' component in the early sections; ensure this is defined before the performance claims.
  3. [Results] Figure or table captions for the imputation results should explicitly list all compared methods and conditions to improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and have revised the manuscript to improve clarity, reproducibility, and transparency.

read point-by-point responses

  1. Referee: [Methods (PMT-constrained co-occurrence knowledge graph and A-TLM description)] The construction of the PMT-constrained co-occurrence knowledge graph is not described in sufficient detail to determine its data sources or independence from the n=946 survey responses used for imputation testing. If edges or co-occurrences are extracted from the same response patterns (even under PMT constraints), retrieval-augmented generation could indirectly access the block-wise missingness patterns being imputed, unlike the classical baselines (MICE+PMM, missForest) that receive no such information. This directly threatens the central claim that A-TLM's RMSE advantage (S4: 1.439) and near-zero bias arise from theory-informed structure rather than leakage. Please specify the exact sources, construction process, and any overlap with the test data.

    Authors: We appreciate the referee's concern regarding potential data leakage, which is essential to substantiate our claims. The PMT-constrained co-occurrence knowledge graph was constructed from external sources, including published literature on Protection Motivation Theory, general co-occurrence statistics from public disaster preparedness datasets, and expert-defined causal relations, all developed independently and prior to the collection or analysis of the n=946 Hurricane Milton survey responses. No edges or co-occurrences were derived from the survey data used in the imputation experiments. We will expand the Methods section with a step-by-step description of the graph construction, explicit listing of all data sources, and verification steps confirming independence from the test data. This revision will demonstrate that the observed advantages in RMSE and bias derive from the theory-informed structure. revision: yes

  2. Referee: [Experimental Setup and LLM Configurations] Exact prompt templates for the seven LLM configurations, the process for integrating the knowledge graph into retrieval, and whether data splits, exclusions, or MNAR simulation parameters were pre-specified versus post-hoc are not provided. Without these, the reported performance numbers (e.g., A-TLM RMSE 1.439 and bias -0.121 under block-wise MNAR) cannot be fully assessed for reproducibility or robustness against implementation choices.

    Authors: We agree that these details are necessary for full reproducibility. We will add the exact prompt templates for all seven LLM configurations to a new appendix. We will also provide a clear description of how the knowledge graph is integrated into the retrieval process. Finally, we will explicitly state in the revised Experimental Setup section that data splits, exclusions, and MNAR simulation parameters were pre-specified in the study protocol prior to any model evaluation or result computation. These additions will enable independent assessment of the reported metrics. revision: yes

  3. Referee: [Results (bias and subgroup analysis)] The manuscript notes that near-zero aggregate bias can mask opposing subgroup errors and proposes subgroup-stratified auditing as a reporting standard, but it is unclear whether this auditing was actually applied to the A-TLM results or only suggested. If not performed, the superiority claim over missForest (largest absolute bias -0.631) remains incomplete, as subgroup-specific errors could undermine the practical utility in disaster preparedness contexts.

    Authors: We thank the referee for this important clarification request. The subgroup-stratified bias auditing was performed on the A-TLM results as part of the analysis to check for masked opposing errors. We will revise the Results section to explicitly document that this auditing was applied, and we will include the subgroup-specific bias and error metrics. This addition will strengthen the evidence for practical utility in disaster preparedness contexts and address the comparison with missForest. revision: yes

Circularity Check

0 steps flagged

No significant circularity; central claims rest on external empirical comparisons

full rationale

The paper evaluates A-TLM imputation performance through direct RMSE and bias metrics against independent classical baselines (IPW/MI, MICE+PMM, missForest) under simulated block-wise MNAR conditions on the n=946 Florida survey. PMT is invoked from established external literature as a constraint on the co-occurrence graph, and the framework compares theory-informed retrieval against unstructured RAG and sequential baselines without reducing the reported gains to internal definitions or self-citations. No load-bearing step equates a prediction to a fitted parameter by construction or imports uniqueness via author-overlapping citations; the derivation chain remains self-contained against the stated external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework relies on Protection Motivation Theory as a domain model and assumes LLM outputs can be reliably constrained by a co-occurrence graph without residual hallucination in the imputation task.

axioms (1)
  • domain assumption Protection Motivation Theory provides a valid causal structure for modeling disaster preparedness survey responses
    Used to organize retrieval and constrain the knowledge graph in the proposed LLM configurations
invented entities (1)
  • Anchored Marginal Theory-Informed LLM (A-TLM) no independent evidence
    purpose: Theory-informed imputation model that integrates PMT structure in a single model call
    New configuration proposed and tested against baselines

pith-pipeline@v0.9.0 · 5838 in / 1264 out tokens · 39693 ms · 2026-05-20T06:26:31.993414+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · 3 internal anchors

  1. [1]

    and Busby, Ethan C

    Argyle, Lisa P. and Busby, Ethan C. and Fulda, Nancy and Gubler, Joshua R. and Rytting, Christopher and Wingate, David , title =. Political Analysis , year =

    work page

  2. [2]

    Journal of Marketing , year =

    Arora, Neeraj and Chakraborty, Inyoung and Nishimura, Yusuke , title =. Journal of Marketing , year =

    work page

  3. [3]

    Sociological Methods & Research , year =

    Ashwin, Julian and Chhabra, Aditi and Rao, Vijayendra , title =. Sociological Methods & Research , year =

    work page

  4. [4]

    Barari, Soubhik and Angbazo, Josephine and Wang, Nuo and Christian, Leah Melani and Dean, Emma and Slowinski, Zachary and Sepulvado, Brandon , title =

    work page

  5. [5]

    Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems , year =

    Bentley, Frank and O'Neill, Kieran and Quehl, Karina and Lottridge, Danielle , title =. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems , year =

    work page 2020

  6. [6]

    and Dorff, Cassy and Kenkel, Brenton and Larson, Jennifer M

    Bisbee, James and Clinton, Joshua D. and Dorff, Cassy and Kenkel, Brenton and Larson, Jennifer M. , title =. Political Analysis , year =

    work page

  7. [7]

    Language Models are Few-Shot Learners

    Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Amodei, Dario , title =. arXiv preprint arXiv:2005.14165 , year =

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  8. [8]

    and Krosnick, Jon A

    Callegaro, Mario and Baker, Reg and Bethlehem, Jelke and Goritz, Anja S. and Krosnick, Jon A. and Lavrakas, Paul J. , title =. Online Panel Research , publisher =. 2014 , pages =

    work page 2014

  9. [9]

    Journal of Marketing , year =

    Chakraborty, Inyoung and Nishimura, Yusuke , title =. Journal of Marketing , year =

    work page

  10. [10]

    Journal of Survey Statistics and Methodology , year =

    Coffey, Spencer , title =. Journal of Survey Statistics and Methodology , year =

    work page

  11. [11]

    , title =

    Couper, Mick P. , title =. Annual Review of Sociology , year =

    work page

  12. [12]

    and Messeri, Lisa , title =

    Crockett, Molly J. and Messeri, Lisa , title =. Trends in Cognitive Sciences , year =

    work page

  13. [13]

    and Christian, Leah Melani , title =

    Dillman, Don A. and Christian, Leah Melani , title =. Field Methods , year =

    work page

  14. [14]

    and Smyth, Jolene D

    Dillman, Don A. and Smyth, Jolene D. and Christian, Leah Melani , title =

    work page

  15. [15]

    and Prentice-Dunn, Steven and Rogers, Ronald W

    Floyd, Donna L. and Prentice-Dunn, Steven and Rogers, Ronald W. , title =. Journal of Applied Social Psychology , year =

    work page

  16. [16]

    2025 , volume =

    Survey and analysis of hallucinations in large language models , journal =. 2025 , volume =

    work page 2025

  17. [17]

    and DeWaard, Jack , title =

    Fussell, Elizabeth and Curtis, Katherine J. and DeWaard, Jack , title =. Population and Environment , year =

    work page

  18. [18]

    Proceedings of the National Academy of Sciences , year =

    Gao, Yang and Lee, Daegon and Burtch, Gordon and Fazelpour, Sina , title =. Proceedings of the National Academy of Sciences , year =

    work page

  19. [19]

    and Couper, Mick P

    Groves, Robert M. and Couper, Mick P. , title =

    work page

  20. [20]

    and Lyberg, Lars , title =

    Groves, Robert M. and Lyberg, Lars , title =. Public Opinion Quarterly , year =

    work page

  21. [21]

    Tracking consumer sentiment versus how consumers are doing based on verified retail purchases , journal =

    Hac. Tracking consumer sentiment versus how consumers are doing based on verified retail purchases , journal =. 2025 , doi =

    work page 2025

  22. [22]

    International Journal of Disaster Risk Reduction , year =

    Hao, Han and Wang, Yan and Kang, Suwan , title =. International Journal of Disaster Risk Reduction , year =

    work page

  23. [23]

    and Liu, Honghu and Kapteyn, Arie , title =

    Hays, Ron D. and Liu, Honghu and Kapteyn, Arie , title =. Behavior Research Methods , year =

    work page

  24. [24]

    https://arxiv.org/abs/2301.07543

    Horton, John J. , title =. arXiv preprint arXiv:2301.07543 , year =

    work page arXiv

  25. [25]

    ACM Transactions on Information Systems , year =

    Huang, Yue and others , title =. ACM Transactions on Information Systems , year =

    work page

  26. [26]

    People and Nature , year =

    Johnson, Christopher and others , title =. People and Nature , year =

    work page

  27. [27]

    and Jansen, Bernard J

    Jung, Soon-gyo and Salminen, Joni and Aldous, Kristen K. and Jansen, Bernard J. , title =. International Journal of Human-Computer Studies , year =

    work page

  28. [28]

    and Mooney, Sean D

    Kaiyrbekov, Kuanysh and Dobbins, Nicolas J. and Mooney, Sean D. , title =. JAMIA Open , year =

    work page

  29. [29]

    The Australian Journal of Emergency Management , year =

    King, David , title =. The Australian Journal of Emergency Management , year =

    work page

  30. [30]

    and Bostrom, Ann and Morss, Rebecca E

    Lazo, Jeffrey K. and Bostrom, Ann and Morss, Rebecca E. and Demuth, Julie L. and Lazrus, Heather , title =. Risk Analysis , year =

    work page

  31. [31]

    and Paulson, Charles , title =

    Leduc, Sylvain and Oliveira, Luiz E. and Paulson, Charles , title =

    work page

  32. [32]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and Kiela, Douwe , title =. arXiv preprint arXiv:2005.11401 , year =

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  33. [33]

    arXiv preprint , year =

    Lin, Yilun and others , title =. arXiv preprint , year =

    work page

  34. [34]

    Proceedings of the National Academy of Sciences , year =

    Loru, Emanuele and Nudo, Jacopo and Di Marco, Nicola and Santirocchi, Alessandro and Atzeni, Riccardo and Cinelli, Matteo and Quattrociocchi, Walter , title =. Proceedings of the National Academy of Sciences , year =

    work page

  35. [35]

    and Craig, S

    Meade, Adam W. and Craig, S. Bartholomew , title =. Psychological Methods , year =

    work page

  36. [37]

    Entropy , year =

    Miranda, Fabio and Balbi, Pier Paolo , title =. Entropy , year =

    work page

  37. [38]

    Electronics , year =

    Naveed, Humza and others , title =. Electronics , year =

    work page

  38. [39]

    Counterfeit judgments in large language models , journal =

    Perc, Matja. Counterfeit judgments in large language models , journal =. 2025 , volume =

    work page 2025

  39. [40]

    Frontiers in Research Metrics and Analytics , year =

    Pinzon-Espitia, Olga Lucia , title =. Frontiers in Research Metrics and Analytics , year =

    work page

  40. [41]

    Humanities and Social Sciences Communications , year =

    Qu, Yiting and Wang, Jia , title =. Humanities and Social Sciences Communications , year =

    work page

  41. [42]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics , year =

    Ravi, Suraj and Rovira, Pol and Acharya, Aditya and Petrov, Slav and Lee, Andrew and Gokul, Hema and Patwary, Mostofa , title =. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics , year =

    work page

  42. [43]

    , title =

    Rubin, Donald B. , title =

    work page

  43. [44]

    and Subrahmanya, Shreyas and Sedoc, Jo

    Salecha, Aadesh and Ireland, Molly E. and Subrahmanya, Shreyas and Sedoc, Jo. Large language models display human-like social desirability biases in. PNAS Nexus , year =

    work page

  44. [45]

    and Rau, Lukas and Schmitt, Bernd , title =

    Sarstedt, Marko and Adler, Stephan J. and Rau, Lukas and Schmitt, Bernd , title =. Psychology & Marketing , year =

    work page

  45. [46]

    SSRN Working Paper No.\ 5976335 , year =

    Saxena, Dhruv , title =. SSRN Working Paper No.\ 5976335 , year =

    work page

  46. [47]

    and White, Ian R

    Seaman, Shaun R. and White, Ian R. and Copas, Andrew J. and Li, Leah , title =. Biometrics , year =

    work page

  47. [48]

    and Peters, Ellen and MacGregor, Donald G

    Slovic, Paul and Finucane, Melissa L. and Peters, Ellen and MacGregor, Donald G. , title =. European Journal of Operational Research , year =

    work page

  48. [49]

    Artificial Intelligence Index Report 2025 , institution =

    work page 2025

  49. [50]

    Stekhoven, Daniel J. and B. Bioinformatics , year =

    work page

  50. [51]

    and Bilgen, Ipek and Dillman, Don A

    Stern, Michael J. and Bilgen, Ipek and Dillman, Don A. , title =. Field Methods , year =

    work page

  51. [52]

    Random silicon sampling: Simulating human sub-population opinion using a large language model based on group-level demographic information

    Sun, Sibo and Lee, Eunsol and Nan, Dong and Zhao, Xiaoqian and Lee, Wonjun and Jansen, Bernard J. and Kim, Jang Hyun , title =. arXiv preprint arXiv:2402.18144 , year =

    work page arXiv

  52. [53]

    Humanities and Social Sciences Communications , year =

    Sun, Sibo and Schoenegger, Philipp and Kapoor, Sandeep and Pan, Liangming and Narayanan, Arvind , title =. Humanities and Social Sciences Communications , year =

    work page

  53. [54]

    arXiv preprint arXiv:2509.07370 , year =

    Tang, Yi and Yang, Yi and Abbasi, Ahmed , title =. arXiv preprint arXiv:2509.07370 , year =

    work page arXiv

  54. [55]

    and Rasinski, Kenneth , title =

    Tourangeau, Roger and Rips, Lance J. and Rasinski, Kenneth , title =

    work page

  55. [56]

    American Community Survey 5-Year Estimates, 2018--2022 , institution =

    work page 2018

  56. [57]

    Journal of Statistical Software , year =

    mice: Multivariate imputation by chained equations in. Journal of Statistical Software , year =

    work page

  57. [58]

    Field Methods , year =

    Wang, Liang and Fan, Xitao , title =. Field Methods , year =

    work page

  58. [59]

    International Journal of Disaster Risk Reduction , year =

    Wang, Yan and Guo, Ziyi and McCarty, Christopher , title =. International Journal of Disaster Risk Reduction , year =

    work page

  59. [60]

    , title =

    Westwood, Sean J. , title =. Proceedings of the National Academy of Sciences , year =

    work page

  60. [61]

    A Survey of Large Language Models

    Zhao, Wayne Xin and Zhou, Kun and Li, Junyi and Tang, Tianyi and Wang, Xiaolei and Hou, Yupeng and Wen, Ji-Rong , title =. arXiv preprint arXiv:2303.18223 , year =

    work page internal anchor Pith review Pith/arXiv arXiv