{"id":"https://openalex.org/W7214966144","doi":"https://doi.org/10.48550/arxiv.2609.35815","title":"How to Run Statistics over LLM Judges and Trust the Results: Calibrated Inference for Small-Sample AI Evaluation with evalstats","display_name":"How to Run Statistics over LLM Judges and Trust the Results: Calibrated Inference for Small-Sample AI Evaluation with evalstats","publication_year":2026,"publication_date":"2026-09-20","ids":{"openalex":"https://openalex.org/W7214966144","doi":"https://doi.org/10.48550/arxiv.2609.35815"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.35815","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35815","pdf_url":"https://arxiv.org/pdf/2609.35815","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.35815","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5054622611","display_name":"Ian Arawjo","orcid":"https://orcid.org/0000-0001-8910-0822","observed_orcids":["https://orcid.org/0000-0001-8910-0822"]},"institutions":[{"id":"https://openalex.org/I70931966","display_name":"Universit\u00e9 de Montr\u00e9al","ror":"https://ror.org/0161xgx34","country_code":"CA","type":"education","lineage":["https://openalex.org/I70931966"]}],"countries":["CA"],"is_corresponding":true,"raw_author_name":"Arawjo, Ian","raw_affiliation_strings":["Department of Computer Science and Operations Research (DIRO ) Universit\u00e9 de Montr\u00e9al Montr\u00e9al , Quebec , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Operations Research (DIRO ) Universit\u00e9 de Montr\u00e9al Montr\u00e9al , Quebec , Canada","institution_ids":["https://openalex.org/I70931966"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5054622611"],"corresponding_institution_ids":["https://openalex.org/I70931966"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T13309","display_name":"Reliability and Agreement in Measurement","score":0.4560999870300293,"subfield":{"id":"https://openalex.org/subfields/1804","display_name":"Statistics, Probability and Uncertainty"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T13309","display_name":"Reliability and Agreement in Measurement","score":0.4560999870300293,"subfield":{"id":"https://openalex.org/subfields/1804","display_name":"Statistics, Probability and Uncertainty"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10206","display_name":"Meta-analysis and systematic reviews","score":0.05209999904036522,"subfield":{"id":"https://openalex.org/subfields/1804","display_name":"Statistics, Probability and Uncertainty"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T13910","display_name":"Computational and Text Analysis Methods","score":0.04190000146627426,"subfield":{"id":"https://openalex.org/subfields/3300","display_name":"General Social Sciences"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/llm-as-a-judge","display_name":"LLM-as-a-judge","score":0.9419999718666077},{"id":"https://openalex.org/keywords/ai-evaluation","display_name":"AI evaluation","score":0.6549999713897705},{"id":"https://openalex.org/keywords/statistical-calibration","display_name":"statistical calibration","score":0.503000020980835},{"id":"https://openalex.org/keywords/prediction-powered-inference","display_name":"prediction-powered inference","score":0.8009999990463257},{"id":"https://openalex.org/keywords/false-positive-risk","display_name":"false positive risk","score":0.7839999794960022},{"id":"https://openalex.org/keywords/confidence-intervals","display_name":"confidence intervals","score":0.6700000166893005},{"id":"https://openalex.org/keywords/hypothesis-testing","display_name":"hypothesis testing","score":0.5730000138282776},{"id":"https://openalex.org/keywords/small-sample-inference","display_name":"small sample inference","score":0.6039999723434448},{"id":"https://openalex.org/keywords/bootstrap-confidence-intervals","display_name":"bootstrap confidence intervals","score":0.6259999871253967},{"id":"https://openalex.org/keywords/interobserver-agreement","display_name":"interobserver agreement","score":0.7049999833106995},{"id":"https://openalex.org/keywords/rank-tests","display_name":"rank tests","score":0.49799999594688416},{"id":"https://openalex.org/keywords/wilcoxon-signed-rank-test","display_name":"Wilcoxon signed-rank test","score":0.7720000147819519},{"id":"https://openalex.org/keywords/mann-whitney-u-test","display_name":"Mann-Whitney test","score":0.9129999876022339},{"id":"https://openalex.org/keywords/false-discovery-rate","display_name":"false discovery rate","score":0.34299999475479126}],"concepts":[{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.7469000220298767},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6179999709129333},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5857999920845032},{"id":"https://openalex.org/C519991488","wikidata":"https://www.wikidata.org/wiki/Q28865","display_name":"Python (programming language)","level":2,"score":0.5120999813079834},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.4959999918937683},{"id":"https://openalex.org/C134261354","wikidata":"https://www.wikidata.org/wiki/Q938438","display_name":"Statistical inference","level":2,"score":0.4603999853134155},{"id":"https://openalex.org/C87007009","wikidata":"https://www.wikidata.org/wiki/Q210832","display_name":"Statistical hypothesis testing","level":2,"score":0.43529999256134033},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.40860000252723694},{"id":"https://openalex.org/C41458344","wikidata":"https://www.wikidata.org/wiki/Q732577","display_name":"Publication","level":2,"score":0.3901999890804291},{"id":"https://openalex.org/C183905921","wikidata":"https://www.wikidata.org/wiki/Q1038757","display_name":"Multiple comparisons problem","level":2,"score":0.38929998874664307},{"id":"https://openalex.org/C96608239","wikidata":"https://www.wikidata.org/wiki/Q1199823","display_name":"Statistical power","level":2,"score":0.3873000144958496},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.37619999051094055},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.373199999332428},{"id":"https://openalex.org/C19499675","wikidata":"https://www.wikidata.org/wiki/Q232207","display_name":"Monte Carlo method","level":2,"score":0.3635999858379364},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33250001072883606},{"id":"https://openalex.org/C44249647","wikidata":"https://www.wikidata.org/wiki/Q208498","display_name":"Confidence interval","level":2,"score":0.3278999924659729},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3181000053882599},{"id":"https://openalex.org/C132964779","wikidata":"https://www.wikidata.org/wiki/Q2110223","display_name":"Raw data","level":2,"score":0.30970001220703125},{"id":"https://openalex.org/C95167961","wikidata":"https://www.wikidata.org/wiki/Q4483495","display_name":"Fiducial inference","level":5,"score":0.3012000024318695},{"id":"https://openalex.org/C158600405","wikidata":"https://www.wikidata.org/wiki/Q5054566","display_name":"Causal inference","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C64869954","wikidata":"https://www.wikidata.org/wiki/Q1859747","display_name":"False positive paradox","level":2,"score":0.2879999876022339},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.2754000127315521},{"id":"https://openalex.org/C207609745","wikidata":"https://www.wikidata.org/wiki/Q4944086","display_name":"Bootstrapping (finance)","level":2,"score":0.275299996137619},{"id":"https://openalex.org/C154606282","wikidata":"https://www.wikidata.org/wiki/Q7048579","display_name":"Nominal level","level":3,"score":0.27129998803138733},{"id":"https://openalex.org/C2779677306","wikidata":"https://www.wikidata.org/wiki/Q2732142","display_name":"Statistician","level":2,"score":0.27090001106262207},{"id":"https://openalex.org/C32792767","wikidata":"https://www.wikidata.org/wiki/Q1370484","display_name":"Summary statistics","level":2,"score":0.2597000002861023},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.2524000108242035}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.35815","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35815","pdf_url":"https://arxiv.org/pdf/2609.35815","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.35815","doi":"https://doi.org/10.48550/arxiv.2609.35815","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.35815","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.35815","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35815","pdf_url":"https://arxiv.org/pdf/2609.35815","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214966144.pdf","grobid_xml":"https://content.openalex.org/works/W7214966144.grobid-xml"},"referenced_works_count":51,"referenced_works":["https://openalex.org/W2127707768","https://openalex.org/W2128213565","https://openalex.org/W2152427206","https://openalex.org/W2398344594","https://openalex.org/W2775663585","https://openalex.org/W2798935874","https://openalex.org/W2939684257","https://openalex.org/W2941486823","https://openalex.org/W2970283086","https://openalex.org/W3133381051","https://openalex.org/W4220893097","https://openalex.org/W4225085525","https://openalex.org/W4232778844","https://openalex.org/W4283170948","https://openalex.org/W4387835444","https://openalex.org/W4388327685","https://openalex.org/W4392223038","https://openalex.org/W4393971165","https://openalex.org/W4396832712","https://openalex.org/W4396833505","https://openalex.org/W4396833703","https://openalex.org/W4399152153","https://openalex.org/W4402671190","https://openalex.org/W4402671800","https://openalex.org/W4403334020","https://openalex.org/W4404345312","https://openalex.org/W4404782677","https://openalex.org/W4407687648","https://openalex.org/W4411505886","https://openalex.org/W4413968549","https://openalex.org/W4414430238","https://openalex.org/W4414564685","https://openalex.org/W4415331408","https://openalex.org/W4415336032","https://openalex.org/W4416036483","https://openalex.org/W4416268504","https://openalex.org/W4416926417","https://openalex.org/W7077142139","https://openalex.org/W7125391887","https://openalex.org/W7128741866","https://openalex.org/W7130415382","https://openalex.org/W7134255296","https://openalex.org/W7139068573","https://openalex.org/W7148271288","https://openalex.org/W7153984434","https://openalex.org/W7154045619","https://openalex.org/W7154163797","https://openalex.org/W7160858152","https://openalex.org/W7196978130","https://openalex.org/W7197048218","https://openalex.org/W7204555598"],"related_works":["https://openalex.org/W7126242407","https://openalex.org/W4416785361","https://openalex.org/W7108335764","https://openalex.org/W7203700165","https://openalex.org/W7121321095","https://openalex.org/W7123312012","https://openalex.org/W7215074674","https://openalex.org/W7171771254","https://openalex.org/W7160858152","https://openalex.org/W7212924736"],"abstract_inverted_index":{"Researchers":[0],"across":[1],"academia":[2],"increasingly":[3],"base":[4],"significance":[5],"claims":[6,27],"on":[7],"LLM":[8,45,77,210],"judge":[9,46,92,211],"scores":[10,47],"and":[11,23,83,94,117,147,167,178,199],"small-sample":[12,174],"AI":[13,175],"evaluations.":[14],"Yet":[15],"without":[16],"well-calibrated":[17],"confidence":[18],"intervals":[19],"(CIs),":[20],"hypothesis":[21,97],"tests,":[22],"judge-bias":[24],"corrections,":[25],"such":[26],"are":[28],"unreliable.":[29],"We":[30,184],"address":[31],"these":[32,186],"issues":[33],"in":[34,202],"several":[35],"contributions.":[36],"First,":[37],"we":[38,80,129,160],"find":[39],"that":[40,150,194],"running":[41],"statistics":[42,75],"over":[43,76],"raw":[44],"leads":[48],"to":[49,73,171,221],"inflated":[50],"false":[51,59],"positives:":[52],"counterintuitively,":[53],"for":[54,85,109,149,163,173],"many":[55],"inter-rater":[56],"agreement":[57],"metrics,":[58],"positive":[60],"risk":[61],"peaks":[62],"at":[63,213,230],"\"almost":[64],"perfect\"":[65],"human-LLM":[66],"agreement.":[67],"To":[68,120],"help":[69],"researchers":[70,180],"understand":[71],"how":[72],"run":[74],"judges":[78],"responsibly,":[79],"present":[81],"guidance":[82],"tooling":[84],"the":[86,104,136,144],"statistical":[87],"analysis":[88],"of":[89],"mixed":[90],"human-AI":[91],"designs,":[93],"implement":[95],"nine":[96],"tests":[98,112],"via":[99],"prediction-powered":[100],"inference":[101],"(PPI),":[102],"including":[103,205],"first":[105],"known":[106],"PPI":[107],"corrections":[108],"four":[110],"rank-based":[111],"(Wilcoxon":[113],"signed-rank,":[114],"Mann-Whitney":[115],"U,":[116],"omnibus":[118],"variants).":[119],"keep":[121],"PPI++":[122],"stable":[123],"with":[124],"small":[125],"human-labeled":[126],"calibration":[127],"sets,":[128],"introduce":[130],"bootstrap-adaptive":[131],"power":[132],"tuning,":[133],"which":[134],"shrinks":[135],"estimated":[137,142],"weight":[138],"toward":[139],"a":[140,208,219,223],"target":[141],"from":[143],"labeled":[145],"data,":[146],"accounts":[148],"weight's":[151],"own":[152],"sampling":[153],"variance.":[154],"Second,":[155],"through":[156],"Monte":[157],"Carlo":[158],"simulations,":[159],"derive":[161],"recommendations":[162,187],"what":[164],"CI,":[165],"p-value,":[166],"FWER":[168],"correction":[169],"methods":[170,197],"use":[172],"evaluations":[176],"(N<100),":[177],"warn":[179],"against":[181],"bootstrap":[182],"CIs.":[183],"package":[185,193],"into":[188],"evalstats,":[189],"an":[190],"open-source":[191],"Python":[192],"selects":[195],"calibrated":[196],"automatically,":[198],"demonstrate":[200],"it":[201],"three":[203],"scenarios,":[204],"one":[206],"where":[207],"real":[209],"validated":[212],"\"substantial":[214],"agreement\"":[215],"would":[216],"have":[217],"led":[218],"researcher":[220],"publish":[222],"spurious":[224],"finding.":[225],"evalstats":[226],"is":[227],"publicly":[228],"available":[229],"https://github.com/ianarawjo/evalstats.":[231]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-01T00:00:00"}