Recent Events and the Coding of Cross-National Indicators

A icle Compa a i e Poli ical S udies 2024, Vol. 57(6) 921–937 © The Au ho (s) 2023 A icle euse guidelines: sagepub.com/jou nals-pe missions DOI: 10.1177/00104140231193006 jou nals.sagepub.com/home/cps Recen E en s and he Coding o C oss-Na ional Indica o s Nils B. Weidmann 1  Abs ac Much esea ch in poli ical science elies on da ase s p oduced by human code s. Many a iables included in hese da ase s a e no based on ob- se able ac s bu a he equi e a conside able le el o human judgmen . This p ojec s udies he ex en o which his judgmen is a ec ed by a ailabili y bias and how i influences he e ospec i e coding o his o ic cases. The analysis uses code -le el da a om he V-Dem p ojec , one o he ew da ase s collec ing and eleasing codings agged wi h imes amps when hey we e p oduced. The esul s show ha ecen d ama ic e en s in a coun y jus p io o he coding ha e a small, bu isible impac on code a ings, bu p ima ily o hose a iables ha a e di ec ly ela ed o he obse ed e en s. The magni ude o his e ec , howe e , is small. This alle ia es conce ns ha p ominen e en s in wo ld poli ics a ound he ime o coding significan ly a ec he eliabili y o c oss-na ional indica o s. Keywo ds quan i a i e me hods, democ a iza ion and egime change, poli ical egimes 1 Depa men o Poli ics and Public Adminis a ion, Uni e si y o Kons anz, Ge many Co esponding Au ho : Nils B. Weidmann, Depa men o Poli ics and Public Adminis a ion, Uni e si y o Kons anz, Uni e si ¨ a ss , 10, Kons anz 78457, Ge many. Email: nils.weidmann@uni-kons anz.de Da a A ailabili y S a emen included a he end o he a icle Kons anze Online-Publika ions-Sys em (KOPS) URL: h p://nbn- esol ing.de/u n:nbn:de:bsz:352-2-1g7cxjpn7ioi44 Much empi ical wo k in poli ical science, and in pa icula in he compa a i e s udy o poli ical egimes, elies on da a p oduced by humans. Using human- coded c oss-na ional da ase s, compa a i e poli ical scien is s can s udy he sp ead (o decline) o democ a ic egimes (Lüh mann & Lindbe g, 2019), moni o he s a e o human igh s wo ldwide (Fa iss, 2014), o gauge he ex en o media eedom ac oss a global sample o coun ies (Kellam & S ein, 2016). Wi hou human-coded c oss-na ional da ase s, much esea ch in compa a i e poli ics, democ a iza ion, and de elopmen would be impossible o conduc . Many o he a iables included in hese da ase s cons i u e subjec i e assessmen s made by he code s. Fo example, when coding he co up ac i i ies o he legisla u e ( a iable 2lgc p ) o he “Va ie ies o Democ acy”da ase (V-Dem, Coppedge e al., 2020), code s need o decide how one could possibly ecognize co up ion o membe s o he legisla u e in e e yday poli ics, and how hese obse ed ou comes ansla e in o he di - e en le els o he coding scale, which anges om “commonly” o “ne e , o ha dly e e .”These codings cons i u e di ficul decisions o he code s: The e a e ew i any p ecise guidelines (i) which empi ical ac s o ake in o accoun o he assessmen and (ii) how o ansla e hem in o quan i a i e sco es, lea ing bo h decisions up o he code . Addi ional complexi y a ises due o ac ha he coding o many indica o s is his o ic and o en imes goes back se e al yea s i no decades. In essence, he coding o hese indica o s is a decision p ocess wi h a conside able le el o unce ain y. Gi en ha a coding p ocess o his kind p o ides ew guidelines o code s, i is likely o be a ec ed by di e en biases in human decision making. Recen wo k is inc easingly ying o unde s and hese cogni i e p ocesses. Colgan (2019), o example, a gues ha human-coded da ase s in IR may ha e an “Ame ican”bias, since hey a e p oduced by code s holding Ame ican alues and iews o he in e na ional sys em. A non e al. (2023) howe e , examining quan i a i e human igh s sco es p oduced by humans, find no e idence o bias. So a he s onges c i ique o human bias in poli ical a ings has been pu o wa d by Li le and Meng (2023), who a gue ha findings o democ a ic decline a e he esul o changes in human pe cep ions bu no in poli ical ins i u ions. Compa ing subjec i e and objec i e democ acy indica o s, hey show ha democ a ic backsliding can be de ec ed in he o me bu no he la e . This pape examines ye ano he ype o bias ha could a ec human code s. Gi en ha coding decisions o his o ic cases a e pa icula ly di ficul , i is easonable o assume ha code s d aw on pa icula cogni i e sho cu s. The a ailabili y bias is one o he mos esea ched among hem. I means ha when con on ed wi h a di ficul decision-making ask, humans use eadily a ailable and closely ela ed in o ma ion o sol e i . In he coding o his o ic, c oss-na ional a iables, his could mean ha human code s a e influenced by ecen d ama ic e en s in he coun y in ques ion, e en i he coding applies o 922 Compa a i e Poli ical S udies 57(6) a ime pe iod many yea s ago. Using ime-s amped code -le el da a om he V-Dem p ojec , he analysis es s how hese ecen e en s a ec code a ings. The esul s show ha he e is some e idence o an a ailabili y bias in code a ings, bu he magni ude emains gene ally small and is unlikely o a ec esea ch done wi h hese indica o s in a majo way. Cogni i e Biases in Human Coding In compa a i e poli ical science esea ch, many human-coded da ase s exis ha co e a a ie y o poli ical a iables. A (non-exhaus i e) lis shows ha hese da a co e many “so ” a iables such as he ule o law (Wo ld Jus ice P ojec , 2020), condi ions o independen media (In e na ional Resea ch & Exchanges Boa d, 2016), p ess eedom (Repo e s wi hou Bo de s, 2020), poli ical igh s and ci il libe ies (F eedom House, 2020a), ins i u ional cha ac e is ics o coun ies (Be ho, 2012), and, o cou se, democ acy (Ma shall e al., 2018;Economis In elligence Uni , 2020;Coppedge e al., 2020). P oducing hese coun y a ings is a di ficul ask o human code s. Some equi e comple ely subjec i e assessmen s, such as, o example, V-Dem’s a iable cap u ing he deg ee o a “ igo ous and impa ial public adminis a ion.” 1 O he s a e a leas in p inciple based on ac ual ou comes o e en s, bu i is e y di ficul o code s o collec all he in o ma ion and/o agg ega e hem o a single coun y a ing acco ding o clea and well-specified coding ules. Fo example, V-Dem codes co up ion in he legisla u e on a 0– 4 scale. The ac s suppo ing his coding may exis , bu hey a e unlikely o be a ailable o he code . 2 Decisions unde unce ain y ha e been he subjec o esea ch o a long ime, mos ly in psychology and ela ed disciplines. T e sky and Kahneman (1974) desc ibe h ee heu is ics ha humans ha e been ound o employ when hey ha e insu ficien knowledge as a basis o a decision. One o hem is he well-known “a ailabili y heu is ic.”A ailabili y means ha when making a decision, humans d aw on hose pieces o in o ma ion ha a e mo e easily a ailable o hem. In his pape , we explo e a ailabili y on a empo al di- mension. When coding a iables o a pa icula coun y o a ime pe iod ha has long passed, i is well possible ha code s a e influenced by ecen , d ama ic e en s in he coun y in ques ion. Wha d ama ic e en s a e mos likely o a ec coding decisions? Exis ing esea ch has p oduced conclusi e e idence o a “nega i i y bias”in news, whe e eade s a e mos likely o pe cei e and emembe nega i e e en s (So oka e al., 2019). In he c oss-na ional s udy o democ acy, whe e code s a e coun ies along a no ma i e dimension be ween “closed au oc acy”and “libe al democ acy,” hese nega i e shi s co espond o se backs away om democ acy, which can mani es hemsel es in mili a y coups o he iolen ep ession o he opposi ion. Consequen ly, i code s a e a ec ed by Weidmann 923 a ailabili y bias, hei e ospec i e a ings o a coun y should be educed i a d ama ic nega i e e en immedia ely p ecedes he coding. A sho example se es o illus a e his. Imagine wo code s a e he le el o a ep ession in a gi en coun y, en yea s ago. This is he ypical coding ask when p oducing c oss-na ional da ase s. The fi s code p oduces his a ing in yea , he second code a yea la e a + 1. In be ween and + 1, he coun y expe iences a la ge wa e o p o es ha is iolen ly ep essed by he go - e nmen , wi h se e al casual ies. A ailabili y bias a ises i hese ecen e en s influence he second code ’s a ing such ha i is lowe han he fi s code ’s a ing, despi e he ac ha bo h code s a e he same his o ic case—bu hey do so a di e en poin s in ime. I code s o compa a i e da ase s a e a ec ed by he a ailabili y heu is ic, he e a e di e en ways in which his can happen. In pa icula , we dis inguish which coding decisions he a ailable in o ma ion (= ecen e en s) will be used o . In a “na ow” e sion o a ailabili y bias, code s use his a ailable in- o ma ion o decisions ha a ec he same ype o phenomenon hey a e supposed o code. The abo e example illus a es his: I a coun y ecen ly expe ienced a d ama ic e en ha clea ly shows a high le el o ep ession (such as a iolen c ackdown agains he opposi ion), code s will use his in o ma ion when a ing he same ou come— he le el o iolen go e nmen ep ession in his coun y. This na ow e sion o a ailabili y bias he e o e sugges s ha ecen d ama ic e en s o go e nmen al ep ession a ec he e ospec i e coding o ep ession- ela ed a iables. Howe e , a ailabili y bias can also play ou in a b oade sense. Ra he han influencing he coding o only hose a iables ha a e di ec ly ela ed o obse ed d ama ic e en s, hese e en s could also a ec o he a iables. Fo example, ha ing obse ed go e nmen iolence agains p o es e s, code s may implici ly downg ade hei no ma i e assessmen o a egime and he e o e assess his egime no jus as ep essi e bu also as “co up ,” “clien elis ic,”“undemoc a ic,”and “illibe al.”I his b oade e sion o a ailabili y holds, we should see an e ec o ecen d ama ic e en s also on a b oad ange o no ma i e a iables, whe e code s change hei e ospec i e assessmen o hese coun ies and a e hem as mo e illibe al a e hese e en s. In he ollowing analysis, we pu he na ow and he b oad e sion o a ailabili y bias o an empi ical es . Resea ch Design The empi ical analysis uses ime-s amped code a ings o a iables ela ed o democ acy and democ a ic ins i u ions. In he ollowing, we desc ibe he da a sou ce and he esea ch design o he analysis. 924 Compa a i e Poli ical S udies 57(6) V-Dem Code -le el Da a The empi ical analysis elies on he code -le el da a om Ve sion 10 o he V-Dem p ojec (Coppedge e al., 2020), a la ge da a collec ion e o ha cap u es many di e en aspec s o “democ acy”and is one o he leading da ase s in he c oss-na ional analysis o egimes. V-Dem dis inguishes be- ween ac ual (Types A and B) and mo e subjec i e (Type C) a iables (Coppedge e al., 2020, p. 28). The la e cons i u e V-Dem’s key con ibu ion and a e he main ocus o his p ojec . Type C a iables cap u e subjec i e a ings o poli ical a iables, p oduced by a numbe o code s wi h pa icula expe ise abou a coun y o egion. Each o hese code s answe s a se o ques ions abou he coun y and ime pe iod hey ha e been assigned. Answe s o he ques ions a e eco ded using di e en scales; ew ques ions ha e a bina y esponse (yes/no), while mos o he s ha e an o dinal scale. The V-Dem code -le el da ase con ains he indi idual a ings p oduced by he code s be o e hey a e p ocessed and agg ega ed u he . In he e sion made a ailable o his esea ch, each coding has a ime s amp associa ed o i , indica ing he yea in which i was p oduced. V-Dem codings a e usually collec ed in Janua y o co e he p e ious (and o en imes also ea lie ) yea s. To maximize he amoun o da a o his analysis while e aining compa a- bili y be ween a iables, only a iables wi h an o dinal scale o 0–4 ( he mos common scale in V-Dem) we e selec ed. 3 Low alues o hese a iables indica e illibe al poli ical p ac ices, and high alues co espond o libe al- democ a ic ea u es o ou comes. The abo emen ioned example o he 2lgc p a iable co e ing co up ion in he legisla u e is an example o his. Figu e 1 illus a es he s uc u e o he code -le el da a om V-Dem. On he imeline along he x-axis, he e is he ac ual case o be coded (he e, he 2lgc p a iable o Egyp in 2016). This a iable is hen coded by h ee code s in ea ly 2018 and wo code s in 2019. We use he e m “coding ime” o deno e he ime (yea ) in which a a iable is coded. In he example, 2lgc p has wo coding imes: 2018 and 2019. Figu e 1. Da a s uc u e o he V-Dem code -le el da a. A a iable ( 2lgc p )is coded o a pa icula case (Egyp in 2016) a wo di e en coding imes (2018 and 2019). Weidmann 925 Uni o Obse a ion In almos all cases, a pa icula a iable o a gi en case is only coded once by a pa icula code , which is why we canno analyze changes wi hin he same code ’s a ings o e ime. Fo ha eason, all a ings o a gi en a iable and case a e a e aged by coding ime; in he abo e example, his would gi e us wo alues o 2lgc p in Egyp 2016: a coding ime 2018, he a e age o code s 14, 16, and 20; and a coding ime 2019, he a e age o code s 34 and 41. O en imes, he e is only a single code a ing o a pa icula coding ime. 4 Fo he analysis, we make use o he ac ha in he V-Dem p ojec , codings o pa icula coun y-yea s a e o en imes gene a ed a di e en poin s in ime. These epea ed obse a ions allow us o examine changes in he a e age coding decisions o e ime, be ween he di e en coding imes. Fo he analysis, we analyze pai s o codings made in consecu i e yea s, in o he wo ds, whe e coding ime 1 and coding ime 2 a e exac ly one yea apa . While o he compa isons a e possible (e.g., we analyze changes in code a ings be ween a gi en yea and h ee yea s la e ), his app oach allows us o be e a ibu e coding changes o he e en s ha happened in be ween he wo coding imes. Fo all pai s p o ided in he code -le el da a, we examine how shi s in he FH sco e in be ween he wo coding imes a ec downg ades in he coding. Measu ing D ama ic E en s The pu pose o he empi ical analysis is o es whe he ecen d ama ic poli ical de elopmen s in a coun y be ween coding ime1 and coding ime2 a ec code s’subjec i e a ings o pas cases. We s a by using di e en e en da ase s o cap u e specific e en s ha indica e a wo sening and inc easingly illibe al poli ical si ua ion. The fi s o hese indica o s is he occu ence o e en s wi h iolen ep ession o p o es in he coded coun y. This a iable is coded om he “Mass Mobiliza ion Da ase ”by Cla k and Regan (2021), selec ing p o es e en s whe e he go e nmen esponded wi h “killings”o “shoo ings.”The second e en -based p edic o is he numbe o coups o coup a emp s, coded om he “Cline Cen e Coup D’´ e a P ojec Da ase ”(Pey on e al., 2020). Bo h ypes o e en s a e usually associa ed wi h high le els o poli ical iolence o supp ess poli ical opposi ion o o ous a democ a ic go e nmen . These e en -based p edic o s, howe e , co e only a small subse o de elopmen s ha indica e whe he a coun y is becoming inc easingly il- libe al. In ac , he e is a b oad ange o e en s ha indica e he de e io a ion o democ acy and could he e o e influence code s’ e ospec i e assessmen . Fo example, he s uggle su ounding Poland’s cons i u ional cou in 2015–2016 was a clea and isible indica o o democ a ic decline bu is 926 Compa a i e Poli ical S udies 57(6) ob iously no cap u ed by he wo e en da ase s. Ra he han expanding he se o e en -based p edic o s (which may no e en be easible due o limi ed da a a ailabili y), we use an agg ega e indica o ha cap u es a b oad ange o shi s away om libe al democ acy: he well-known F eedom House “F eedom in he Wo ld” a ing (FH hence o h). The FH sco es quan i y a coun y’s le el o poli ical igh s and ci il libe ies wi h an annual sco e be ween 0 and 100 (F eedom House, 2020a). This sco e is compu ed as he sum o se e al cons i uen indica o s, cap u ing he elec o al p ocess, poli ical plu alism and pa icipa ion, he unc ioning o go e nmen , he eedom o exp ession and belie , associa ional and o ganiza ional igh s, he ule o law, as well as pe sonal au onomy and indi idual igh s. In many cases, F eedom House a ings change slowly. These changes a e unlikely o be isible o code s and he e o e no expec ed o influence hei coding as hey do no cons i u e “d ama ic”e en s. We he e o e selec majo changes om F eedom House in wo ways. Fi s , we selec cases whe e a coun y was downg aded by a leas 5 poin s in he FH scale wi hin a single yea , o cap u e majo poli ical shi s away om democ acy. These majo shi s happen a ely. Since we can only exploi a ia ion in he coding imes, we can only use changes du ing he yea s 2016–2020. Du ing his ime, only abou 3% o he coun y-yea s expe ienced FH d ops o 5 o mo e. A second way o iden i y majo downg ades is by using he h ee FH s a us ca ego ies. FH classifies a coun y as “ ee,”“pa ly ee,”o “no ee,”depending on he alues o he unde lying poli ical igh s and ci il libe ies sco es (F eedom House, 2020b). A s a us downg ade happens i a coun y d ops om “ ee” o “pa ly ee”o om “pa ly ee” o “no ee.”These s a us downg ades a e majo e en s ha a e discussed in he annual FH news elease, so hey a e likely o be isible o he code s. In he analysis below, we es whe he om any coding yea o he nex , a d op in he FH sco e o a leas 5 poin s o a s a us downg ade leads o an adjus men o he code a ings downwa d. Resul s In he empi ical analysis, we assess he ex en o which ecen , d ama ic e en s in a coun y a ec he code s’ e ospec i e a ings. To do so, we p oceed in a s epwise ashion. We s a wi h a pooled analysis using se s o V-Dem a iables, be o e es ing each a iable independen ly. The analysis ocuses on he change in he code a ings be ween wo consecu i e yea s. Mo e p ecisely, he dependen a iable is he di e ence o he a e age V-Dem code a ing be ween coding ime1 and coding ime2. O e all, code assessmen s a e ela i ely s able. Ou o he 117,817 obse a ions in he main da ase , in 31,482 cases (abou 27%) he coding emains unchanged, while upg ades (44,288, abou 37%) and downg ades (42,047, abou 36%) occu almos a he same a e. The dis ibu ion o changes in code a ings om one yea o he Weidmann 927 nex is isualized in Figu e 2. I shows ha while he as majo i y o a ings do no change (la ge peak a he cen e ), downg ades and upg ades a e simila in magni ude and almos pe ec ly balanced. Appendix A2 p esen s summa y s a is ics a he le el o he a iables, dis inguishing be ween a ia ion be- ween he indi idual cases ( he s anda d de ia ion o he a e age code a ings o each coun y/yea ) and he a e age a ia ion wi hin a case ( he mean ac oss he s anda d de ia ions o he code a ings o each coun y/yea ). Fo he analysis, we use OLS models wi h he fi s di e ence o he code a ings ( he a ing a coding ime2 minus he a ing a coding ime1) as he dependen a iable. The pooled models include fixed e ec s o he di e en V-Dem a iables, since he scaling o each o hem could en ail pa icula ly low o high changes om yea o yea . In addi ion, he models clus e he s anda d e o s a he le el o cases (coun y/yea s), since hese obse a ions a e subjec o he same poli ical de elopmen s and he e o e no independen . Changes in Code Ra ings In line wi h he heo e ical discussion abo e, we fi s es a na ow e sion o he a ailabili y heu is ic, whe e code s a e influenced by ecen e en s in he coun y o be coded bu use his in o ma ion only o code a iables ha a e closely ela ed o hese e en s. The fi s eg ession uses he e en -based indica o s o d ama ic iolen e en s ( ep essed p o es and coups) o see i hese e en s a ec he coding o V-Dem a iables ela ed o iolence and ep ession. In V-Dem, he e a e fi e o hese a iables; hese include 2cl o ( eedom om o u e), 2clkill ( eedom om poli ical killings), 2cs ep ss (go e nmen al ep ession o ci il socie y o ganiza ions), 2cs lg ep (go - e nmen al ep ession o eligious o ganiza ions), and 2meha j n (physical ha assmen o jou nalis s). The fi s wo models in Table 1 es how he iolen ep ession o p o es (Model 1) and he occu ence o a coup (Model 2) in be ween he wo coding imes a ec he changes in he code a ings. In line wi h ou expec a ion, bo h Figu e 2. Dis ibu ion o di e ences in he code a ings. 928 Compa a i e Poli ical S udies 57(6) ecei e a nega i e e ec , which indica es ha he occu ence o hese e en s is associa ed wi h a d op in he code a ings. The magni ude o he e ec is lowe o ep essed p o es , whe e an e en o his kind is associa ed wi h a d op in he code assessmen s by only abou .15 on he 0–4 scale. The e ec o a coup is conside able, indica ing ha e ospec i e a ings o ep ession- ela ed V-Dem a iables a e on a e age .6 poin s lowe on he 0–4 scale i a coun y expe ienced a coup. In Models 3–6inTable 1, we expand he analysis such ha i includes all V-Dem a iables as well as he al e na i e indica o s o d ama ic e en s based on F eedom House. Models 3 and 4 es ima e he impac o he e en -based p edic o s ( ep essed p o es and coups) on changes in he a ings o all a iables, Models 5 and 6 use he occu ence o an FH downg ade o a leas 5 poin s (Model 5) o a s a us downg ade (Model 6) as p edic o s. The co- e ficien s in hese models a e much smalle h oughou . Only ep essed p o es is significan , bu again in he di ec ion we hypo hesized. The es ima e shows ha he occu ence o physical ep ession agains p o es e s leads o a Table 1. E ec o Recen E en s in Coded Coun y on Changes in Code Ra ings. OLS Models Wi h Va iable FEs and S anda d E o s Clus e ed by Case (Coun y/ Yea ). Dependen a iable ΔCode a ing Rep ession- ela ed a iables All a iables (1) (2) (3) (4) (5) (6) Rep essed p o es .148*** (.030) .075*** (.016) Coup o coup a emp .616*** (.079) .072 (.046) FH downg ade ≥4 .023 (.028) FH s a us downg ade .083 (.059) V-Dem a iable FEs Yes Yes Yes Yes Yes Yes Clus e ed SEs (c /yea ) Yes Yes Yes Yes Yes Yes Obse a ions 13,431 13,431 117,817 117,817 117,817 117,817 Adjus ed R2 .008 .008 .002 .002 .002 .002 No e.*p< .05; **p< .01; ***p< .001. Weidmann 929 Da a A ailabili y S a emen Replica ion ma e ials and code can be ound in he Compa a i e Poli ical S udies Da a e se a h ps://doi.o g/10.7910/DVN/FKQI6M. Supplemen al Ma e ial Supplemen al ma e ial o his a icle is a ailable online. No es 1. Va iable 2cl spc , ques ion: “A e public o ficials igo ous and impa ial in he pe o mance o hei du ies?”(Coppedge e al., 2020, 164). 2. Va iable 2lgc p , ques ion: “Do membe s o he legisla u e abuse hei posi ion o financial gain?”(Coppedge e al., 2020, 138). 3. To ensu e compa ibili y wi h o he annual codings, a iables ha do no ollow an annual coding pa e n (e.g., hose ela ed o elec ions) we e also excluded. We also exclude a iables ha do no cons i u e no ma i e assessmen s, such as he ques ion abou he dominan chambe in bicame al legisla u es. See he appendix o he comple e lis o a iables included. 4. This esul s in (non-sys ema ic) measu emen e o , which is unlikely o bias he esul s we ob ain om he agg ega ed codings. Re e ences A non, D., Haschke, P., & Pa k, B. (2023). The igh accoun ing o w ongs: Examining empo al changes o human igh s moni o ing and epo ing. B i ish Jou nal o Poli ical Science,53(1), 163–182. h ps://doi.o g/10.1017/s0007123421000661 Be ho, F. (2012). “P esen a ion o he ins i u ional p ofiles da abase 2012.”h p:// www.cepii. /ins i u ions/doc/IPD_2012_cahie s-2013-03_EN.pd Cla k, D., & Regan, P. (2021). Mass mobiliza ion p o es da a. Ha a d Da a e se. h ps://massmobiliza ion.gi hub.io Colgan, J. D. (2019). Ame ican bias in global secu i y s udies da a. Jou nal o Global Secu i y S udies,4(3), 358–371. h ps://doi.o g/10.1093/jogss/ogz030 Coppedge, M., Ge ing, J., Lindbe g, S. I., Jan, T., Al man, D., Be nha d, M., S e en Fish, M., Glynn, A., Allen, H., Anna, L., Ma qua d , K. L., Kelly, M. M., Pax on, P., Pems ein, D., Sigman, R., S a on, J., Wilson, S., Co nell, A., Alizada, N., & Medziho sky, J. (2020). V-dem da ase V10 codebook. Va ie ies o democ acy (V-dem) p ojec . Economis In elligence Uni . (2020). “Democ acy index 2019: A yea o democ a ic se backs and popula p o es .”Economis In elligence Uni . h ps://www.eiu. com/ opic/democ acy-index Fa iss, C. J. (2014). Respec o human igh s has imp o ed o e ime: Modeling he changing s anda d o accoun abili y. Ame ican Poli ical Science Re iew,108(2), 297–318. h ps://doi.o g/10.1017/s0003055414000070 F eedom House. (2020a). F eedom House Coun y Ra ings.h p://www. eedomhouse.o g. 936 Compa a i e Poli ical S udies 57(6) F eedom House. (2020b). F eedom in he wo ld 2020 me hodology. Me hodo- logical Documen a ion. h ps:// eedomhouse.o g/ epo s/ eedom-wo ld/ eedom-wo ld- esea ch-me hodology In e na ional Resea ch & ExchangesBoa d. (2016). Media sus ainabili y index (MSI). h ps://www.i ex.o g/ esou ce/media-sus ainabili y-index-msi Kellam, M., & S ein, E. A. (2016). Silencing c i ics: Why and how p esiden s es ic media eedom in democ acies. Compa a i e Poli ical S udies,49(1), 36–77. h ps://doi.o g/10.1177/0010414015592644 Li le, A., & Meng, A. (2023). Subjec i e and objec i e measu emen o democ a ic backsliding. SSRN. h ps://www.ss n.com/abs ac =4327307 Lüh mann, A., & Lindbe g, S. I. (2019). A hi d wa e o au oc a iza ion is he e: Wha is new abou i ? Democ a iza ion,26(7), 1095–1113. h ps://doi.o g/10.1080/ 13510347.2019.1582029 Ma shall, M. G., Ted, R. G., & Kei h, J. (2018). Poli y IV p ojec : Poli ical egime cha ac e is ics and ansi ions, 1800-2018. Technical Repo . h ps://www. sys emicpeace.o g/insc /p4manual 2018.pd Pems ein, D., Ma qua d , K. M., Tzelgo , E., Wang, Y., Medziho sky, J., K usell, J., Mi i, F., & Johannes, V. R. (2020). The V-dem measu emen model: La en a iable analysis o c oss-na ional and c oss- empo al expe -coded da a. Technical epo V-Dem P ojec . h ps:// -dem.ne /s a ic/websi e/files/wp/ wp_21_5 h.pd . Pey on, B., Joseph, B., Shalmon, D., Ma in, M., & Bonagu o, J. (2020). Cline cen e coup D’´ e a p ojec da ase cline cen e o ad anced social esea ch. V.2.0.0 No embe 16. Uni e si y o Illinois U bana-Champaign. h ps://doi.o g/ 10.13012/B2IDB-9651987_V3 Repo e s wi hou Bo de s (2020). Wo ld p ess eedom index.h ps:// s .o g/en/ anking. So oka, S., Fou nie , P., & Ni , L. (2019). C oss-na ional e idence o a nega i i y bias in psychophysiological eac ions o news. P oceedings o he Na ional Academy o Sciences o he Uni ed S a es o Ame ica,116(38), 18888–18892. h ps://doi. o g/10.1073/pnas.1908369116 T e sky, A., & Kahneman, D. (1974). Judgmen unde unce ain y: Heu is ics and biases. Science,185(4157), 1124–1131. h ps://doi.o g/10.1126/science.185.4157.1124 Wo ld Jus ice P ojec . (2020). Rule o law index. Wo ld Jus ice P ojec . h ps:// wo ldjus icep ojec .o g/ou -wo k/ esea ch-and-da a/wjp- ule-law-index-2020 Au ho Biog aphy Nils Weidmann is P o esso o Poli ical Science and head o he “Com- munica ion, Ne wo ks and Con en ion”Resea ch G oup a he Uni e si y o Kons anz. His esea ch in e es s include poli ical p o es and iolen conflic , he poli ical impac s o new communica ion echnology, and poli ical me hodology. Weidmann 937