Buchne , Ma in; Rehm, Mi iam
Wo king Pape
Ene gy po e y and heal h: Mic o-le el e idence om
Ge many
i so wo king pape , No. 48
P o ided in Coope a ion wi h:
Uni e si y o Duisbu g-Essen, Ins i u e o Socioeconomics (i so)
Sugges ed Ci a ion: Buchne , Ma in; Rehm, Mi iam (2025) : Ene gy po e y and heal h: Mic o-le el
e idence om Ge many, i so wo king pape , No. 48, Uni e si y o Duisbu g-Essen, Ins i u e o
Socio-Economics (i so), Duisbu g
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/313638
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by/4.0
uni-due.de/soziooekonomie/wp
i so wo king pape
Ma in Buchne
Mi iam Rehm
Ene gy Po e y and Heal h:
Mic o-Le el E idence om
Ge many
2025 no.48
Ene gy Po e y and Heal h:
Mic o-Le el E idence om Ge many∗
Ma in Buchne †Mi iam Rehm‡
Oc obe 2024
∗We hank B uno B inkmeie o excellen assis ance in he p epa a ion o his pape , and pa icipan s
o he I SO esea ch semina , he “Poli ical Economy o Inequali y” summe school, and he EuHEA PhD
Con e ence o commen s and eedback. Na u ally, all emaining e o s a e ou own. Ma in Buchne
g a e ully acknowledges PhD unding om he Hans-Böckle Founda ion.
†RWI – Leibniz Ins i u e o Economic Resea ch & Uni e si y o Duisbu g-Essen, Ins i u e o Socio-
Economics, ma in.buc[email p o ec ed]
‡Uni e si y o Duisbu g-Essen, Ins i u e o Socio-Economics, [email p o ec ed]
Abs ac
This pape aims o unde s and he heal h e ec s o ene gy po e y in Ge many using
SOEP panel da a om 2010 o 2020. Linea p obabili y models and ixed e ec s
o de ed logi models e eal a consis en ly nega i e ela ionship o h ee expendi u es-
based ene gy po e y indica o s wi h gene al heal h. The associa ion is s onge o
he subjec i e ene gy po e y me ic: membe s o households unable o keep he home
com o ably wa m due o inancial easons ha e an abou 3.23 p.p. lowe p obabili y
o being in a leas sa is ac o y heal h. In es iga ing po en ial channels shows ha
men al heal h is consis en ly nega i ely linked o ou ene gy po e y me ics, while
physical heal h is weakly associa ed wi h ene gy po e y in Ge many, wi h he excep-
ion o doc o isi s. Finally, by ins umen ing ene gy po e y wi h da a on ene gy
p ice indices and ma ching ene gy cos s o he hea ing sys ems used by households,
we show ha li ing in a household ha expe iences a ansi ion o ene gy po e y
due o ising ene gy p ices is also linked o a lowe likelihood o being in good heal h.
Keywo ds: ene gy po e y, heal h, ixed e ec s o de ed logi models, Ge many
JEL Codes: I10, I32, Q41
1
1 In oduc ion
Ene gy po e y o households can make hei membe s sick. When households s uggle o
a ain an adequa e le el o ene gy se ices, leading o ene gy o uel po e y (Boa dman,
1991; Bouza o ski, 2014), i can esul in inancial s ain on he one hand, and o insu icien
hea ing (and cooling) o li ing qua e s on he o he hand (Da illas e al., 2022). Such
ci cums ances can p ecipi a e a ious physical heal h p oblems, including inc eased isk o
hype ension, in lamma ion, ca dio ascula diseases, h ombosis, and espi a o y illnesses
in child en, as well as men al heal h issues (Galle ani e al., 2004; Fa es, 2013; Balles e os-
A jona e al., 2022). Ene gy po e y is hus a policy conce n bo h na ionally, highligh ed by
ad ocacy g oups in he Uni ed Kingdom since he 1970s, and a he Eu opean le el, whe e
he Eu opean Commission has issued a ecommenda ion on add essing ene gy po e y
(Eu opean Commission, 2020). In Ge many, ecen es ima es sugges a p e alence o ene gy
po e y o abou 6.6% (Des a is, 2023). Wi h sho - un spikes in ene gy p ices and long-
e m ends s emming om he Ge man ene gy ansi ion, he inancial bu den o ene gy
cos s on households is likely o inc ease. This pape aims o in es iga e whe he a obus
link exis s be ween ene gy po e y and men al and physical heal h in Ge many.
While he e is ample esea ch on low- and middle-income coun ies (Bane jee e al.,
2021; Jayasinghe e al., 2021; Nawaz, 2021; Nie e al., 2021; Pan e al., 2021), ecen s udies
ha e unco e ed obus ela ionships be ween ene gy po e y and heal h in se e al high-
income coun ies. No ably, ou key s udies using su ey panel da a o F ance (Kahouli,
2020; Baudu e al., 2020, EU-SILC), Aus alia (Chu chill and Smy h, 2021, HILDA),
and he UK (Da illas e al., 2022, UKHLS) ha e de eloped a amewo k o es ima ing
he causal e ec o a ious ene gy po e y measu es on objec i e and sel - a ed heal h
ou comes. Sel - epo ed heal h se es as a common ou come a iable ac oss hese s udies,
wi h Chu chill and Smy h (2021) employing a composi e indica o o subjec i e heal h and
Da illas e al. (2022) in addi ion accessing objec i e heal h da a om blood samples. The
explana o y a iable, ene gy po e y, is measu ed h ough bo h objec i e and subjec i e
dimensions, wi h all h ee s udies employing he low income-high cos measu e1(Hills,
2012) as an objec i e measu e, along wi h subjec i e assessmen s ega ding he abili y
o adequa ely hea he home. Chu chill and Smy h (2021) and Da illas e al. (2022)
also inco po a e a composi e o objec i e and subjec i e measu es o ene gy po e y. To
add ess endogenei y conce ns, ene gy po e y is ins umen ed wi h egional ene gy p ices
in all h ee s udies, gi en hei co ela ion wi h he po en ially endogenous explana o y
a iable, ene gy po e y, and hei p esumed lack o di ec linkage o heal h ou comes.
1This measu e indica es ha he sha e o ene gy expendi u es exceeds 10% o equi alised disposable
household income.
2
Addi ionally, s anda d con ols include indi idual and dwelling cha ac e is ics, as well as
clima e ac o s.
This esea ch yields aluable insigh s in o he ela ionship be ween ene gy po e y and
heal h. In addi ion o he obus headline co ela ion o ene gy po e y and ad e se heal h
ou comes (Champagne e al., 2023), i shows consis en indings e en when going in o mo e
de ail. Fo ins ance, subjec i ely measu ed ene gy po e y exhibi s a s onge nega i e
e ec on heal h compa ed o measu es assessing he inancial bu den o ene gy cos s, such
as he " en pe cen ule" (ene gy cos s amoun ing o mo e han en pe cen o he household
budge ) o he "low income-high cos " indica o (a composi e indica o based on i s wo
namesake concep s). Howe e , some subs an ial a ia ions pe sis . No ably, ins umen al
a iable (IV) e ec sizes a y conside ably ac oss s udies (e.g., Baudu e al., 2020; Kahouli,
2020; Da illas e al., 2022; Chu chill and Smy h, 2021), emphasizing he signi icance o he
conc e e IV speci ica ion bu also o na ional ins i u ional cha ac e is ics.
Ge many is cha ac e ised by a mo e decen alised ins i u ional amewo k, including
he social policy egime, han F ance o he Uni ed Kingdom, and by a less empe a e cli-
ma e han Aus alia o F ance. Thus, his s udy seeks o asce ain whe he a obus link
be ween ene gy po e y and heal h ou comes exis s in he Ge man con ex .2Le e aging
ep esen a i e panel da a om he Ge man Socio-Economic Panel (SOEP) spanning om
2010 o 2020, ou ou come a iable heal h is measu ed subjec i ely using a sel -assessed in-
dica o . Sel - a ed heal h me ics a e ecognized as eliable p oxies o ac ual heal h s a us,
combining bo h men al and physical heal h conside a ions (Schni ke and Bacak, 2014),
and a e commonly employed in empi ical heal h economics (e.g., Kuehnle and Wunde ,
2017) including s udies on ene gy po e y (Llo ca e al., 2020). Fo ene gy po e y mea-
su emen , we employ h ee expendi u e-based me ics ha conside he a io o household
income o ene gy expendi u es alongside a subjec i e indica o , which cons i u e he mos
widely used me ics in ene gy po e y esea ch (B abo-Ca ala e al., 2024). The la e
may be e cap u e unde consump ion o ene gy due o inancial cons ain s and i a oids
labelling high-income, high-ene gy use households as ene gy-poo (Thema and Vondung,
2020; D esche and Janzen, 2021).
Ou es ima ion s a egy aims o es ablish a obus link be ween ene gy po e y and
heal h ou comes. Ini ially, we i a linea p obabili y model be o e u ning o panel-da a
o de ed logi models wi h ixed e ec s using he blow-up and clus e (BUC) es ima o
p oposed by Bae schmann e al. (2020) and Bae schmann e al. (2015). This app oach
2Reibling and Ju z (2017) p esen a i s quan i a i e indica ion o a nega i e co ela ion be ween en-
e gy po e y and men al heal h. Due o da a es ic ions, howe e , his s udy ocuses exclusi ely on hea ing
expendi u es (wi hou aking elec ici y cos s in o accoun ), uses a single expendi u e-based measu e o
ene gy po e y, and canno assess causali y.
3
accommoda es he o de ed scaling o he dependen a iable (sel - a ed heal h) and exploi s
he panel s uc u e o he da a by con olling o po en ial unobse ed indi idual ime-
in a ian con ounde s. Addi ionally, we add ess possible biases in wo-way ixed e ec s
models wi h s agge ed and in e mi en ea men by implemen ing he inno a i e Fixed
E ec s coun e ac ual es ima o p oposed by Liu e al. (2024). Ac oss all h ee app oaches,
we consis en ly obse e a s a is ically signi ican nega i e associa ion be ween expendi u e-
based ene gy po e y indica o s and o e all heal h. No ably, bo h he linea p obabili y
model and he ixed e ec s o de ed logi model show ha he link be ween subjec i e
ene gy po e y and heal h is especially s ong.
Subsequen ly, we in esiga e po en ial channels h ough which ene gy po e y impac s
physical and men al heal h ou comes. Ou indings sugges ha he nega i e associa ion
p ima ily s ems om de e io a ions in men al a he han physical heal h in Ge many, as
e idenced by composi e indica o s and indi idual a iables.
Fu he mo e, we adop an IV app oach o add ess he po en ial endogenei y be ween
heal h ou comes and ene gy po e y. To his end, we ins umen ene gy po e y using
da a on p ice indices o oil, gas, dis ic hea , solid uels, and elec ici y. Ou da a pe mi
us o iden i y he p ima y ene gy sou ce households use o hea ing, which is c ucial o
ins umen ing ene gy po e y since le e aging he a ia ion in ene gy p ices is he leading
app oach in ins umen ing ene gy po e y. They a e gene ally assumed o be ele an ,
since p ice inc eases a e likely o be co ela ed wi h ene gy po e y. The exclusion e-
s ic ion is ha de o sa is y (Kahouli, 2020; Chu chill and Smy h, 2021): An inc ease in
ene gy p ices can a ec heal h s a us beyond inducing ene gy po e y. Fo ins ance, i may
p omp households o educe di ec expendi u es on heal h- ela ed p oduc s and se ices,
such as gym membe ships o heal hy die s, in esponse o he p ice su ge. Howe e , his
subs i u ion e ec is less o a conce n i ene gy expendi u e changes a e small ela i e o
o al expendi u es o i heal h expendi u es a e p ice inelas ic (Da illas e al., 2022). The
esul s o he IV poin o a s ong nega i e link be ween ene gy po e y and heal h, and
a e hus in line bo h wi h ou mul i a ia e indings and he li e a u e.
Finally, we ex ensi ely es he obus ness o hese esul s by checking ha ou indings
a e no in luenced by ea u es o he Ge man wel a e s a e, which co e s hea ing expen-
di u es wi h some social ans e s. Addi ionally, we es ic he sample o di ec su ey
esponden s and non-COVID yea s, and exclude o e -sampled mig an s.
The es o his pape is s uc u ed as ollows: Sec ion 2 ou lines ou me hodology and
he empi ical s a egy, while Sec ion 3 desc ibes he da a and a iables. Sec ion 4 p esen s
ou indings, commencing wi h he linea p obabili y models in Sec ion 4.1, ollowed by
ixed-e ec s o de ed logi models in Sec ion 4.2, and wo-way ixed e ec s coun e ac ual
4
models in Sec ion 4.3. Sec ion 4.4 explo es po en ial channels, and Sec ion 4.5 add esses
endogenei y wi h he IV app oach. Sec ion 5 examines he obus ness o ou esul s, and
Sec ion 6 concludes.
2 Empi ical S a egy
To assess he ela ionship be ween ene gy po e y and an indi idual’s o e all heal h s a-
us, we i s u ilize a Linea P obabili y Model (LPM) o consis ency wi h he exis ing
li e a u e. LPMs o e wo key ad an ages: i s , hei esul s a o d a simple and in ui i e
in e p e a ion; and second, hey a e well-es ablished in he li e a u e and hus acili a e
compa abili y wi h o he s udies, such as Kahouli (2020). The es ima ed equa ion is as
ollows:
good_heal hi =β1EPi +X
n
βnXn,i +ϑi+γ +εi ,(1)
whe e good_heal hi is he sel - a ed heal h o esponden ia ime , which is assigned a
alue o 1 i he esponden ’s sel - a ed heal h s a us is “ e y good”, “good” o “sa is ac o y”
and 0 i he esponse is “poo ” o “bad”. EP is he espec i e indica o o ene gy po e y
and Xis a ec o o n indi idual ime- a ying obse ed con ol a iables. Indi idual ixed
e ec s a e cap u ed by ϑi, while γ deno es wa e dummies co e ing ime ixed e ec s.
We p oceed by employing ixed-e ec s o de ed logis ic eg ession models, ou p e e ed
app oach. These models a e ad an ageous as hey a oid he need o dicho omize he de-
penden a iable, sel - a ed heal h, which is measu ed on an o de ed ca ego ical scale, while
s ill le e aging he panel da a s uc u e o accoun o unobse ed ime- and indi idual-
in a ian ac o s. Howe e , in o de o es ima e he ixed e ec s consis en ly despi e he
so-called inciden al pa ame e s p oblem (Lancas e , 2000), Bae schmann e al. (2015) in-
oduce a me hod le e aging he Condi ional Maximum Likelihood (CML) es ima o . Fo
o de ed ca ego ial dependen a iables, he Blow-Up and Clus e es ima o (BUC) com-
bines he „in o ma ion o he CML es ima o s ob ained om dicho omizing samples a
di e en cu o poin s“ (Bae schmann e al., 2020) by eplacing he sample wi h copies o
i sel and applying he CML es ima o o he en i e enla ged sample.3The BUC-app oach
no only allows o a consis en es ima ion o ixed e ec s o de ed logi models, bu i also
does no ely on he assump ion ha he h esholds a e cons an ac oss indi iduals, in
con as o s anda d o de ed logi models o c oss-sec ional da a.
3Since he copies o he same uni s a e no independen o each o he , s anda d e o s a e clus e ed a
he indi idual le el.
5
We ope a ionalize he ixed-e ec s o de ed logis ic models by employing sel - a ed
heal h as dependen a iable, which is scaled om 1 (bad) o 5 ( e y good heal h s a-
us). The app oach can handle he same independen a iables, including ene gy po e y,
indi idual and yea ixed e ec s, and con ols, as he linea p obabili y model.
Al hough he ixed e ec s o de ed logi model is ou p e e ed app oach gi en he da a
s uc u e, i has been shown ha TWFE es ima es may su e om bias s emming om
weigh ing issues, pa icula ly when ea men e ec s exhibi empo al a ia ion in s ag-
ge ed ea men designs (Chaisema in and d’Haul oeuille, 2023; Goodman-Bacon, 2021;
Chaisema in and d’Haul oeuille, 2020). Ou panel da a is suscep ible o his conce n,
gi en ha indi iduals esiding in households expe ience ansi ions in and ou o ene gy
po e y h oughou ou analysis pe iod.4To add ess his issue, we employ he inno a-
i e Fixed E ec s coun e ac ual es ima o (FEc ) p oposed by Liu e al. (2024), which
calcula es he a e age ea men e ec on he ea ed by di ec ly impu ing coun e ac ual
ou comes o ea ed obse a ions. This app oach ea s ea ed obse a ions as missing
du ing modelling and es ima es he coun e ac ual ou come as a weigh ed sum o all un-
ea ed obse a ions. This me hod compa es he obse ed ou come o ea ed obse a ions
wi h he p edic ed coun e ac ual o each ma ched pai , hus emo ing biases caused by
imp ope weigh ing ha a ec con en ional TWFE models (Liu e al., 2024). We employ
he FEc es ima o o ou expendi u e-based ene gy po e y indica o s, es o he pa al-
lel ends assump ion, and pe o m placebo es s by emo ing p e- ea men obse a ions
used a he modelling s age.
Despi e con olling o ime-in a ian he e ogenei y and nume ous po en ial ime- a ian
con ounding a iables in all discussed models, as well as conduc ing es s o p e- ends and
placebo ea men s in he FEc models, conce ns ega ding iden i ica ion may s ill a ise i
ene gy po e y is co ela ed wi h he e o e m. The e a e a leas h ee possible sou ces
o endogenei y: Fi s , unobse ed ime- a ying con ounding a iables no accoun ed o
in he model speci ica ions (see Equa ion 1) could bias he es ima ion o he eg ession
coe icien s. Fo ins ance, indi idual expec a ions conce ning job loss o income may in-
luence bo h heal h and ene gy po e y (Kahouli, 2020). A second po en ial sou ce o
endogenei y is e e se causa ion. Fo ins ance, indi iduals in poo heal h migh ea n less,
he eby inc easing he likelihood o being in a s a e o ene gy po e y, which could bias
he eg ession es ima es upwa d. Finally, endogenei y may s em om measu emen e o .
Su ey esponden s migh no accu a ely ecall hei ene gy bills, leading o sys ema ic
o e - o unde es ima ion o hei annual ene gy expendi u es.
4Figu e A5 in he Appendix p o ides a isual ep esen a ion o indi iduals’ ea men s a us in ou
sample o e ime.
6
om he Fede al S a is ical O ice (S a is isches Bundesam , 2024). Howe e , i is impo -
an o no e ha due o a me hodological change in he calcula ion o he p ice indices,
da a o he dis ic hea p ice index begins in 2015.
We inco po a e one-yea lagged ene gy p ice index alues o ou Ins umen al Va iable
(IV) es ima es, d i en by wo conside a ions: Fi s , in he SOEP da ase , homeowne s yp-
ically epo ene gy expendi u es om he p eceding yea , while en e s, who epo cu en
expendi u es, a e in luenced by u ili y bills based on p io ene gy p ices. Second, gi en
ha SOEP in e iews p edominan ly occu du ing he sp ing mon hs14, ene gy p ices om
he p e ious yea e ain g ea e salience when esponden s epo hei ene gy expendi-
u e. Consequen ly, ou ins umen comp ises 49 unique alues, e lec ing a ia ions ac oss
5 hea ing ypes and 11 ime pe iods (see Figu e 3).
In assigning hea ing sys ems o households, we le e age a unique ea u e o he SOEP
da ase . In 2015 and 2020, households we e su eyed abou hei expendi u e on hea ing
pe ene gy sou ce15. When hea ing cos s a e epo ed o a single ene gy sou ce, we assign
ha speci ic hea ing sys em o he household o he espec i e yea . In ins ances whe e
expenses a e incu ed o mul iple ene gy sou ces, we p io i ise based on he ollowing
hie a chy, e lec ing he ela i e equency in Ge many: gas > oil > dis ic hea ing >
elec ici y > solid uels > o he (Des a is, 2022). We ex end his assigned hea ing sys em
o all yea s p eceding and succeeding he obse ed yea , p o ided he household does no
epo a mo e. We assume ha as long as he e is no mo e, he household’s hea ing
sys em emains unchanged. Howe e , i he hea ing sys ems di e be ween 2015 and 2020,
indi iduals esiding in hose households a e excluded om ou ins umen al a iable (IV)
analysis, as we canno de e mine he iming o he swi ch. Ou IV analysis hus only
includes obse a ions om households ha consis en ly u ilize he same ene gy sou ce o
hea ing o e ime.
The summa y s a is ics o he educed subsample, consis ing o indi iduals esiding
in households whe e we ha e success ully assigned a hea ing sys em and ob ained he
co esponding p ice index om he p e ious yea , a e p esen ed in Table A2. Rema kably,
no subs an ial de ia ions a e obse ed be ween his educed sample and ou main sample
conce ning he dependen a iable good_heal h and he ene gy po e y indica o s.
14See Figu e A4 in he Appendix, which shows ha app oxima ely 50% o in e iews ake place in
Feb ua y and Ma ch, and o e 75% occu in he i s hal o he yea .
15Expendi u es can be epo ed ac oss a ious ca ego ies, including oil, gas, dis ic hea ing, elec ici y,
en i onmen al, pelle s, coal, biomass, liquid gas, and sola . Fo he analysis p esen ed in his pape , we
agg ega e hese ca ego ies in o gas, oil, dis ic hea ing, elec ici y, solid uels, and o he , as p ice da a a e
a ailable only o he i s i e o hese ene gy sou ces.
13
Figu e 3: Ene gy p ice indices
This igu e shows he ene gy p ice indices o gas, oil, dis ic hea , elec ici y, and solid ules, indexed o 2020 as 100. We
employ he one-yea lag o he p ice indices in ou ins umen al a iable (IV) models. Da a: S a is isches Bundesam (2024).
4 Resul s
This sec ion i s p esen s he esul s o ene gy po e y and heal h h ough s anda d linea
p obabili y models, o de ed logis ic ixed-e ec s, and wo-way ixed e ec s coun e ac ual
models. I hen mo es o examining he e idence o po en ial channels be ween he wo
and a emp s o add ess hei po en ial endogenei y using an IV app oach. I concludes
wi h obus ness checks.
4.1 Linea P obabili y Models
Table 2 shows linea p obabili y models ocusing on he h eshold be ween e y good
o sa is ac o y heal h e sus poo o bad heal h s a us. The esul s indica e a nega i e
co ela ion be ween heal h s a us and ene gy po e y ac oss all ou ene gy po e y indi-
ca o s (columns 1-4). No ably, as shown in column 4, he e ec size o he subjec i e
ene gy po e y indica o is subs an ially la ge compa ed o he expendi u e-based indica-
o s. When ene gy po e y is subjec i ely measu ed, i is associa ed wi h a 3.2 pe cen age
14
poin s lowe p obabili y o epo ing a leas a sa is ac o y heal h s a us.
All ou speci ica ions in Table 3 con ol o o he ac o s po en ially ela ed o heal h as
discussed in Sec ion 3, including loga i hmized and equi alized household income, income
po e y, socio-economic cha ac e is ics such as age and age squa ed, educa ion, labou
o ce s a us (including being egis e ed as unemployed), ma i al s a us, and he numbe
o dependan s li ing in he household. Mos ele an o ou esea ch ques ion a e he
co a ia es cap u ing income po e y and unemploymen , bo h o which a e s a is ically
signi ican ly associa ed wi h wo se heal h.
4.2 O de ed Logis ic Fixed-E ec s Models
Nex , we examine whe he he nega i e associa ion be ween ene gy po e y and heal h
pe sis s when conside ing he o de ed ca ego ical na u e o he dependen a iable and he
panel da a s uc u e. Fixed-e ec s o de ed logis ic es ima es o he ela ionship be ween
ene gy po e y and gene al heal h a iables a e p esen ed in odds a ios in Table 3. Simila
o he linea p obabili y model, we inco po a e he ull se o con ols. The exponen ia ed
coe icien s o all h ee expendi u e-based ene gy po e y indica o s (columns 1-3) a e
s a is ically signi ican a leas a he 5%-le el and smalle han one, sugges ing ha ene gy
po e y is associa ed wi h a dec ease in he odds a io o being in be e heal h ca ego ies.
Fo ins ance, he odds a io o 0.945 in column 1 indica es ha li ing in a household
classi ied as ene gy poo by he Ten Pe cen Rule ( p ) dec eases he odds a io o being in
a be e heal h ca ego y by abou 5.5%. In o he wo ds, indi iduals in Ten Pe cen Rule
ene gy-poo households ha e 0.945 imes he odds o being in be e heal h ca ego ies han
hose in non-ene gy-poo households. Simila ly, he odds a ios o he Two Times Median
Sha e o Income (m wo) and he Low Income High Cos (lihc) indica o s a e 0.943 and
0.919, espec i ely. As shown in column 4, he odds a io o he subjec i e ene gy po e y
indica o is subs an ially la ge han hose o he expendi u e-based indica o s. I s poin
es ima e o 0.818 indica es ha li ing in a household unable o keep he home com o ably
wa m due o inancial easons dec eases he odds a io o being in a be e heal h ca ego y
by abou 18,2%.
Table 4 p esen s he ma ginal e ec s a he sample a e age, which may be mo e in u-
i i e o in e p e .16 These indica e ha being ene gy poo acco ding o ou ou me ics
(columns 1-4) is associa ed wi h lowe p obabili ies o being in he wo bes heal h ca -
ego ies (lines 4-5) and highe p obabili ies o being in he h ee lowe heal h ca ego ies
16Ma ginal e ec s a he a e age use he ela i e equencies o he co esponding ca ego ies o he
ou come a iable in he es ima ion sample o compu e he sample a e age. No e ha he ma ginal e ec s
a he sample a e age di e om he ma ginal e ec s a he a e age o he eg esso s and also di e om
he a e age ma ginal e ec . Fo a mo e de ailed discussion, see Bae schmann e al. (2020).
15
Table 2: Ene gy po e y and heal h – linea p obabili y model
(1) (2) (3) (4)
Good heal h Good heal h Good heal h Good heal h
Ten pe cen ule ( p ) -0.00828∗∗∗
[0.00268]
Two imes median sha e (m wo) -0.00879∗∗
[0.00370]
Low income high cos s (lihc) -0.00892∗∗
[0.00396]
subjec i e -0.0323∗∗
[0.0134]
Income po e y -0.0108∗∗∗ -0.0112∗∗∗ -0.0114∗∗∗ -0.0119
[0.00375] [0.00373] [0.00372] [0.00723]
Log o eq. HH income 0.00270 0.00366 0.00448 0.00443
[0.00364] [0.00361] [0.00357] [0.00681]
Age 0.00215∗0.00218∗0.00213∗0.0179∗∗∗
[0.00111] [0.00111] [0.00111] [0.00346]
Age squa ed -0.0000812∗∗∗ -0.0000810∗∗∗ -0.0000808∗∗∗ -0.000209∗∗∗
[0.0000107] [0.0000107] [0.0000107] [0.0000335]
Regis e ed Unemployed -0.0551∗∗∗ -0.0554∗∗∗ -0.0554∗∗∗ -0.0542∗∗∗
[0.00497] [0.00497] [0.00497] [0.0100]
No employed -0.0202∗∗∗ -0.0203∗∗∗ -0.0202∗∗∗ -0.00912
[0.00399] [0.00399] [0.00399] [0.00836]
In Educa ion / App en ice / Communi y Se ice -0.00177 -0.00188 -0.00181 -0.0116
[0.00469] [0.00469] [0.00469] [0.00925]
Pensione 0.0125∗∗ 0.0124∗∗ 0.0125∗∗ 0.0129
[0.00569] [0.00569] [0.00569] [0.0111]
High School 0.00338 0.00362 0.00364 -0.0299
[0.00868] [0.00867] [0.00867] [0.0182]
Mo e han High School 0.00380 0.00404 0.00393 -0.0470∗
[0.0116] [0.0116] [0.0116] [0.0245]
Ma ied 0.00182 0.00180 0.00180 0.00165
[0.00550] [0.00550] [0.00550] [0.0137]
Sepa a ed o di o ced 0.0118 0.0115 0.0115 0.0166
[0.00832] [0.00832] [0.00832] [0.0196]
Widowed -0.00288 -0.00294 -0.00360 0.00641
[0.0143] [0.0144] [0.0144] [0.0342]
Numbe o dependan s in hh 0.000885 0.000942 0.00101 -0.00453
[0.00185] [0.00185] [0.00185] [0.00441]
Cons an 0.926∗∗∗ 0.916∗∗∗ 0.912∗∗∗ 0.512∗∗∗
[0.0341] [0.0340] [0.0338] [0.0945]
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 255684 255684 255684 86211
This able shows he es ima es o a linea p obabili y model o he ou ene gy po e y indica o s on he p obabili y o being
in good heal h (dicho omized as 1 o e y good, good o sa is ac o y, and 0 o poo o bad). The e e ence ca ego y o labou
o ce s a us is (sel -)employed, o educa ion is less han high school, and o ma i al s a us is single. Clus e obus s anda d
e o s in b acke s.
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
16
Table 3: O de ed logis ic eg ession
(1) (2) (3) (4)
Sel a ed Sel a ed Sel a ed Sel a ed
heal h heal h heal h heal h
Ten pe cen ule ( p ) 0.945∗∗∗
[0.0176]
Two imes median sha e (m wo) 0.943∗∗
[0.0230]
Low income high cos s (lihc) 0.919∗∗∗
[0.0235]
subjec i e 0.818∗∗
[0.0744]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
Obse a ions ( o al) 255684 255684 255684 86211
Obse a ions (w. a ia ion in ou come) 206662 206662 206662 52671
Panel uni s (w. a ia ion in ou come) 31797 31797 31797 15581
This able shows he exponen ia ed coe icien s o a ixed-e ec s o de ed logis ic eg ession o he
ou ene gy po e y indica o s on sel - a ed heal h measu ed on a 5-poin scale wi h 1 = bad and
5 = e y good. Clus e obus s anda d e o s in b acke s. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
(lines 1-3). The esul s sugges ha i a household alls in o Ten Pe cen Rule ene gy
po e y, he e is a dec ease o 1.4 pe cen age poin s in he p obabili y o being in good o
e y good heal h o i s membe s. These esul s a e consis en ac oss all ou indica o s;
o subjec i e ene gy po e y, he associa ion again appea s o be he s onges , wi h a
educ ion o app oxima ely 5 pe cen age poin s in he p obabili y o being in good o e y
good heal h.
4.3 Two-way Fixed E ec s Coun e ac ual Models
As a hi d modeling app oach, we employ ixed e ec s coun e ac ual models (FEc ) o
coun e ac po en ial weigh ing issues s emming om ou s agge ed and in e mi en ea -
men . The dependen a iable is a dummy a iable ha cap u es whe he indi iduals
epo sa is ac o y, good, o e y good heal h, as opposed o poo o bad heal h. The main
explana o y a iables a e he expendi u e-based indica o s, as he subjec i e indica o o
ene gy po e y is a ailable o oo ew pe iods o adequa ely es o pa allel ends.
Figu es A6 o A8 in he Appendix p o ide a isual assessmen o he pa allel ends
assump ion, complemen ed by an F- es o ze o esidual a e ages in he p e ea men
pe iods, whe e a highe F- es p- alue indica es a be e i o p e- end analysis. The
ATT plo s and he p- alues o he F- Tes collec i ely indica e no subs an ial p esence
17
Table 4: Ma ginal e ec s
(1) (2) (3) (4)
p m wo lihc subjec i e
sel - a ed heal h
1 0.00192∗∗∗ 0.00200∗∗ 0.00285∗∗∗ 0.00712∗∗
[0.000632] [0.000827] [0.000868] [0.00323]
2 0.00660∗∗∗ 0.00691∗∗ 0.00984∗∗∗ 0.0250∗∗
[0.00218] [0.00285] [0.00299] [0.0113]
3 0.00556∗∗∗ 0.00582∗∗ 0.00829∗∗∗ 0.0176∗∗
[0.00184] [0.00240] [0.00252] [0.00798]
4 -0.00910∗∗∗ -0.00953∗∗ -0.0136∗∗∗ -0.0317∗∗
[0.00300] [0.00393] [0.00413] [0.0144]
5 -0.00498∗∗∗ -0.00521∗∗ -0.00742∗∗∗ -0.0180∗∗
[0.00164] [0.00215] [0.00226] [0.00815]
This able shows he ma ginal e ec s o he ou ene gy po e y indica o s in a
ixed-e ec o de ed logis ic eg ession, calcula ed a he sample a e age o he
dependen a iable (sel - a ed heal h on a 5-poin scale wi h 1 = bad and 5 = e y
good). The epo ed s anda d e o s a e o e ec s a he sample a e age and no
o e ec s a he popula ion a e age. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
o p e ea men di e en ial ends o mos models while he p- alue o he - es o he
placebo es in he Low Income High Cos model is bo de line s a is ically signi ican a
he 10% le el.17 Fu he mo e, we conduc a placebo es whe ein all obse a ions om
pe iods -2 o 0 ela i e o he ea men iming a e excluded. Ins ead, he un ea ed
ou comes o hese omi ed pe iods a e p edic ed using a model ained wi h he emaining
un ea ed obse a ions (Liu e al., 2024). The ou comes a e depic ed in Figu es A9-A11 in
he Appendix: In he placebo es s, we canno ejec he null hypo hesis o an ATT = 0 o
he Two Times Median Sha e o Income and Low Income High Cos models, wi h p- alues
o 0.967 and 0.825, espec i ely. This ou come suppo s he alidi y o he unde lying
assump ions o hese models.
17This is no ewo hy, especially conside ing ha he F-Tes is pa icula ly sensi i e o de ia ions om
ze o: „[W]hen he sample size is la ge, a small con ounde (o a ew ou lie s) ha only con ibu es o a
neglec able amoun o bias in he causal es ima es will almos always cause ejec ion o he null hypo hesis
o join ze o means using he F es “. (Liu e al., 2024).
18
Table 5: Fixed e ec s coun e ac ual models
EP Indica o ATT N sd lowe CI uppe CI p- alue
p -0.01008 28502 0.003204 -0.01529 -0.005 0.002
m wo -0.01268 13842 0.004397 -0.0197 -0.0051 0.004
lihc -0.01024 11966 0.004931 -0.01844 -0.00223 0.038
This able shows ixed e ec s coun e ac ual es ima ions. S anda d e o s a e ob ained h ough non-
pa ame ic boo s ap p ocedu es (500 boo s ap uns). Uni s mus ha e a minimum o 1 obse ed pe iod
unde con ol o be conside ed. Da a: SOEP (2022)
The a e age ea men e ec s o he ea ed (ATT) a e p esen ed in Table 5. We
obse e simila bu somewha la ge nega i e ATTs compa ed o he coe icien s ob ained
om he linea p obabili y model. Speci ically, being in ene gy po e y is associa ed wi h
a dec ease in he p obabili y o being in good o e y good heal h by app oxima ely 1
pe cen age poin o he Ten Pe cen Rule and he Low Income High Cos indica o , and
abou 1.3 pe cen ages poin s o he Two Times Median indica o .
Ou esul s hus a demons a e a obus nega i e associa ion be ween ene gy po e y
and heal h in Ge many, aligning wi h indings om he li e a u e on o he high-income
coun ies. As discussed in Sec ion 1, Kahouli (2020), Baudu e al. (2020), and Chu chill
and Smy h (2021), and Da illas e al. (2022) es ablish his ela ionship o F ance, Aus-
alia, and he UK. Mo eo e , Kahouli (2020) and Chu chill and Smy h (2021) also obse e
s onge e ec s o consensual-based (i.e., subjec i e) indica o s compa ed o expendi u e-
based me ics. In he nex sec ion, we explo e po en ial channels h ough which ene gy
po e y may a ec heal h in Ge many.
4.4 Po en ial Channels
We now u n o explo ing he po en ial channels h ough which ene gy po e y may impac
heal h. Table 6 p esen s he esul s o ixed-e ec s eg essions o ou ull model, u ilizing
men al heal h (le -hand side panel) and physical heal h ( igh -hand side panel) summa y
scales as dependen a iables.18 The associa ions wi h ou ene gy po e y me ics exhibi
s a k a ia ions: Men al heal h is nega i ely associa ed wi h ene gy po e y, s a is ically
signi ican a leas a he 5% le el o all ou expendi u e-based me ics. Howe e , he e
appea s o be no such associa ion o physical heal h.
While he magni ude o he coe icien s o he men al heal h summa y scale associ-
18Recall ha he dependen a iable is a ailable biennially, leading o a educed numbe o obse a ions
o 121,227. Consequen ly, we canno u ilize he subjec i e ene gy po e y me ic due o i s limi ed o e lap
wi h he in e al censo ed physical and men al heal h summa y scales.
19
a ed wi h he h ee ene gy po e y me ics may no be di ec ly in e p e able due o he
composi e na u e o he ou come a iable, he coe icien s exhibi consis en sizes. This
associa ion pe sis s e en a e ex ensi e con ols o po en ial con ounding ac o s such as
income po e y and unemploymen , mi o ing he indings om he p e ious Sec ion.
Table 6: Men al and physical heal h summa y scales – SF12-ques ionai e
Men al Summa y Scale Physical Summa y Scale
Ten pe cen ule ( p ) -0.354∗∗∗ -0.0438
[0.108] [0.0868]
Two imes median sha e (m wo) -0.401∗∗∗ 0.0460
[0.146] [0.118]
Low income high cos s (lihc) -0.322∗∗ 0.241∗
[0.162] [0.130]
Socio-econ. con ols Yes Yes Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes Yes Yes
Indi idual ixed e ec s Yes Yes Yes Yes Yes Yes
Yea ixed e ec s Yes Yes Yes Yes Yes Yes
n 121227 121227 121227 121227 121227 121227
This able shows ixed-e ec s eg essions o he h ee objec i e ene gy po e y indica o s on
men al and on physical heal h summa y scales ( anging om 0 o 100). Due o da a a ailabili y,
his eg ession co e s e e y o he yea (2010, 2012, ..., 2020). Indi idual clus e ed obus
s anda d e o s in pa en heses. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
Nex , we explo e hese wo channels u he by employing indi idual i ems ela ed o
men al and physical heal h. Ou da a pe mi s us o subs i u e he dependen a iable wi h
h ee “ha d” indica o s o physical heal h and i e “so ” indica o s o men al heal h: he
loga i hmized numbe o isi s o a doc o ’s o ice in he p e ious yea , he numbe o s ays
a a hospi al in he p e ious yea , and he loga i hmized numbe o days o wo k sick o
physical heal h; and li e sa is ac ion, he equency o being happy o sad in he las ou
weeks, heal h sa is ac ion, and conce ns abou heal h o men al heal h.
The esul s a e epo ed in he Appendix. They ein o ce he conclusions d awn om
he men al and physical heal h summa y scales: Physical heal h demons a es a weak, i
any, connec ion o ene gy po e y a e adjus ing o income po e y and unemploymen .
While he numbe o doc o isi s (Table A3) displays some co ela ion wi h ene gy po e y,
coe icien s o hospi al s ays (Table A4) and he numbe o days o wo k (Table A5) a e
s a is ically insigni ican .
Con e sely, indica o s ela ed o men al heal h consis en ly subs an ia e a link wi h
ene gy po e y e en a e con olling o o he ac o s. Li e sa is ac ion (Table A6), heal h
sa is ac ion ( o a lesse ex en , Table A7), and he equency o being happy (Table A8)
a e nega i ely associa ed wi h all ou ene gy po e y indica o s, while he equency o
being sad (Table A9) and, less consis en ly, conce ns abou one’s heal h (Table A10) end
20
o inc ease wi h ene gy po e y.
This obus ela ionship hus suppo s he hypo hesis ha , a leas wi hin he Ge man
con ex , he nega i e heal h impac s o ene gy po e y may p ima ily o igina e om i s
e ec s on men al well-being. Howe e , as discussed in Sec ion 2, a ious po en ial sou ces
o endogenei y o ene gy po e y o heal h necessi a e cau ion in in e p e ing hese indings.
Consequen ly, he nex sec ion in es iga es whe he hese indings emain obus in an IV
analysis.
4.5 Add essing Endogenei y
To assess whe he he nega i e associa ion be ween ene gy po e y and heal h pe sis s while
add essing po en ial endogenei y, we employ 2SLS eg essions, ins umen ing o ene gy
po e y wi h household-speci ic ene gy p ices using ene gy p ice da a. The esul s o he
IV ixed-e ec s eg ession a e p esen ed in Table 7. We inco po a e he same co a ia es
as in ou main speci ica ion, including a dummy o income po e y. The Kleibe gen-
Paap Wald F-s a is ic o he i s s age sugges s ha he ins umen is no weak o he
Ten Pe cen Rule and Two Times Median Sha e o Income indica o s (S ock and Yogo,
2005). Howe e , o he Low Income High Cos indica o , he F-s a is ic is ma ginally
la ge han 10, indica ing only a weak co ela ion be ween household-speci ic ene gy p ices
and his o m o ene gy po e y. No ably, only he F-s a is ic o he Ten Pe cen Rule
indica o exceeds he h eshold o 104.7, as p oposed by Lee e al. (2022). One po en ial
explana ion is ha he Ten Pe cen Rule indica o does no ely on a e e ence popula ion.
Consequen ly, when ene gy p ices ise, he median ene gy expendi u e in he popula ion
is likely o ise oo, somewha o se ing he p obabili y o becoming Two Times Median
Sha e o Income o Low Income High Cos ene gy poo . The e o e, we p ima ily ocus on
he esul s de i ed om he Ten Pe cen Rule indica o .
In he second s age o he 2SLS eg ession, we ind a nega i e and s a is ically signi ican
coe icien o bo h he Ten Pe cen Rule and he Two Times Median Sha e o Income
me ic. Li ing in a household ha expe iences a ansi ion o Ten Pe cen Rule ene gy
po e y due o ising ene gy p ices is linked o a decline o being in good heal h o abou 15.5
pe cen age poin s. Fo Two Times Median Sha e o Income ene gy po e y, his associa ion
is e en s onge a abou 37.2 pe cen age poin s. Howe e , gi en he s ong co ela ion
in he i s s age o he 2SLS es ima es, we conside he Ten Pe cen Rule indica o ou
p e e ed measu e o ene gy po e y he e.
21
Table 7: Ene gy po e y and heal h (IV es ima es)
p m wo lihc
Second s age
Ene gy po e y -0.1546∗-0.3717∗-0.7364
[0.0876] [0.2162] [0.4656)]
Obse a ions 146,331 146,331 146,331
Fi s s age
Ene gy p ices (CPI) 0.0008∗∗∗ 0.0004∗∗∗ 0.0002∗∗∗
[0.0000771] [0.0000561] [0.0000533]
Kleibe gen-Paap Wald F s a is ic 120.49 39.376 11.114
This able shows he ins umen al a iable es ima es o he h ee objec i e en-
e gy po e y indica o s on he p obabili y o being in good heal h (dicho omized
as 1 o e y good, good o sa is ac o y, and 0 o poo o bad). All eg essions
include he usual co a ia es. Indi idual clus e ed obus s anda d e o s in
pa en heses. Da a: SOEP (2022).
*p < 0.1,** p < 0.05,*** p < 0.01
The endogenei y bias-co ec ed es ima es de i ed om he 2SLS model appea sub-
s an ial, a pa e n consis en wi h much o he li e a u e u ilizing ene gy p ices as ins u-
men s (Chu chill and Smy h (2021); Chu chill and Smy h (2020) Kahouli (2020); P akash
and Munyanyi (2021)). Howe e , his may indica e linge ing conce ns ega ding he ex-
clusion es ic ion, u ging cau ion agains an o e ly con iden in e p e a ion o causali y.
None heless, he sign o he coe icien s associa ed wi h ene gy po e y in he 2SLS es i-
ma es emains consis en wi h he baseline speci ica ion, ein o cing ou o e all conclusion
ha ene gy po e y signi ican ly and de imen ally impac s indi iduals’ heal h.
5 Robus ness Checks
We es he obus ness o ou indings h ough se e al s a egies. Fi s , we add ess a
speci ici y o he Ge man wel a e sys em by excluding households po en ially ecei ing
ans e s co e ing ene gy cos s. Second, we es ic he analysis o household heads, since
hey espond o he SOEP household ques ionnai e which con ains he ene gy expendi u e-
ela ed ques ions. Thi d, we omi he yea 2020 om ou analysis o mi iga e po en ial
dis o ions a ibu able o he COVID-19 pandemic. Finally, we elimina e esponden s
who we e pa o he mig a ion samples in he SOEP da a o ensu e he consis ency and
eliabili y o ou esul s.
Fi s , ce ain households in Ge many ecei e ans e s co e ing housing cos s, includ-
ing hea ing (“A bei slosengeld” o “Wohngeld”). Al hough ou da a do no allow o he
22
A Appendix
A.1 Da a Desc ip ion
Table A1: Be ween and wi hin a ia ion o ene gy po e y me ics
Va iable Mean S d. de . Obse a ions
p o e all .2177453 .4127142 N = 255684
be ween .3631767 n = 56263
wi hin .25968 T-ba = 4.54444
m wo o e all .0852263 .2792187 N = 255684
be ween .2509484 n = 56263
wi hin .1881695 T-ba = 4.54444
lihc o e all .0713967 .2574869 N = 255684
be ween .2253253 n = 56263
wi hin .1758772 T-ba = 4.54444
subjec i e o e all .0158796 .1250106 N = 86211
be ween .1039507 n = 31941
wi hin .0797766 T-ba = 2.69907
This able shows he a ia ion o he ou ene gy po e y measu es be ween
and wi hin indi iduals. Da a: SOEP (2022).
Figu e A1: Venn diag ams
Objec i e indica o (2010-2020)
Objec i e and subjec i e indica o (2016-
2019)
Venn diag ams illus a ing he in e sec ion o ene gy po e y indica o s wi hin ou pooled sample (le panel: 2010-2020;
igh panel: 2016-2019.). Reading guide o le panel: 13,187 obse a ions (indi idual-yea combina ions) in ou pooled
sample a e iden i ied as ene gy poo by bo h he p and he m wo indica o s. Addi ionally, 199,921 obse a ions a e no
classi ied as ene gy poo acco ding o any o ou objec i e ene gy po e y me ics. Da a: SOEP (2022).
29
Figu e A4: Mon h o SOEP in e iew
Dis ibu ion o pe son in e iews in ou sample by mon h and yea . 141 obse a ions a e excluded due o missing
in o ma ion on he mon h he in e iew ook place. Da a: SOEP (2022).
30
Table A2: Summa y S a is ics - IV Sample
Mean SD Min Max N
sel - a ed heal h 3.39 0.95 1.00 5.00 146331
good heal h 0.83 0.38 0.00 1.00 146331
p 0.20 0.40 0.00 1.00 146331
m wo 0.08 0.27 0.00 1.00 146331
lihc 0.07 0.25 0.00 1.00 146331
subjec i e 0.02 0.12 0.00 1.00 60417
gas 0.56 0.50 0.00 1.00 146331
oil 0.28 0.45 0.00 1.00 146331
dis ic hea 0.08 0.27 0.00 1.00 146331
elec ici y 0.04 0.20 0.00 1.00 146331
solid uels 0.05 0.21 0.00 1.00 146331
Summa y s a is ics o he educed IV-Sample. Da a: SOEP (2022).
31
A.2 Two-way Fixed E ec s Coun e ac ual (FEc )
Figu e A5: T ea men S a us Ten Pe cen Rule Ene gy Po e y
Visualiza ion o he ea men s a us o ene gy po e y (acco ding o he p , m wo,
lihc, and subjec i e indica o s, espec i ely) o all indi iduals in he sample. Uni s a e
so ed based on he iming o ecei ing he ea men o isualiza ion pu poses. Da a:
SOEP (2022).
32
Figu e A6: FEc – p Figu e A7: FEc – m wo
Figu e A8: FEc – lihc
Fixed E ec s Coun e ac ual es ima ions. S anda d e o s a e ob ained h ough non-pa ame ic boo s ap p ocedu es (500
boo s ap uns). Uni s mus ha e a minimum o 1 obse ed pe iod unde con ol o be conside ed. Plo is limi ed o pe iods
whe e he numbe o ea ed obse a ions exceeds 20 pe cen o he la ges numbe o ea ed obse a ions in a pe iod
(de aul is 30 pe cen ; we had o lowe his h eshold in o de o ha e a leas 3 p e- ea men pe iods). We es o he
p esence o p e ea men di e en ial ends by using a a ian o he F-Tes buil -in he ec -package ha es s o ze o
esidual a e ages in he p e ea men pe iods (la ge F- es p- alues sugges be e p e- end i ing). Da a: SOEP (2022).
33
Figu e A9: FEc placebo es – p Figu e A10: FEc placebo es – m wo
Figu e A11: FEc placebo es – lihc
The placebo es is pe o med by emo ing all obse a ions om he pe iods -2 o 0 ela i e o ea men iming o model
i ing. The p- alue indica es whe he he es ima ed ATT in his ange is signi ican ly di e en om ze o. S anda d e o s
a e ob ained h ough non-pa ame ic boo s ap p ocedu es (500 uns). Da a: SOEP (2022).
34
A.3 Robus ness Checks
Table A3: Fixed-e ec s eg ession - log numbe o doc o isi s
(1) (2) (3) (4)
Log doc o Log doc o Log doc o Log doc o
isi s isi s isi s isi s
Ten pe cen ule ( p ) 0.0124∗∗
[0.00515]
Two imes median sha e (m wo) 0.0116∗
[0.00700]
Low income high cos s (lihc) 0.00540
[0.00740]
subjec i e 0.0510∗∗
[0.0256]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 248713 248713 248713 85812
Fixed-e ec s eg ession. Dependen a iable: Log o numbe o isi s o a doc o ’s o ice in he
p e ious h ee mon hs. Clus e obus s anda d e o s in b acke s. All eg essions include he
usual co a ia es.
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
Table A4: Fixed-e ec s eg ession - hospi al s ays
(1) (2) (3) (4)
hospi al s ays hospi al s ays hospi al s ays hospi al s ays
Ten pe cen ule ( p ) -0.00291
[0.00309]
Two imes median sha e (m wo) 0.00333
[0.00446]
Low income high cos s (lihc) -0.0000771
[0.00445]
subjec i e 0.00159
[0.0148]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 196707 196707 196707 73713
Fixed-e ec s eg ession. Dependen a iable: Hospi al s ays in su ey yea (0=no s ay, 1=a leas one
hospi al s ay). Clus e obus s anda d e o s in b acke s. All eg essions include he usual co a ia es.
Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
Table A5: Fixed-e ec s eg ession - numbe o days o wo k
(1) (2) (3) (4)
Log days Log days Log days Log days
o wo k o wo k o wo k o wo k
Ten pe cen ule ( p ) -0.0245
[0.0156]
Two imes median sha e (m wo) 0.0259
[0.0247]
Low income high cos s (lihc) 0.0145
[0.0248]
subjec i e 0.0320
[0.0682]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 123568 123568 123568 48916
Fixed-e ec s eg ession. Dependen a iable: Log o numbe o days o wo k sick in he
espec i e su ey yea . Clus e obus s anda d e o s in b acke s. All eg essions include
he usual co a ia es. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
36
Table A6: Fixed-e ec s eg ession - li e sa is ac ion
(1) (2) (3) (4)
Li e sa is . Li e sa is. Li e sa is . Li e sa is .
Ten pe cen ule ( p ) -0.0655∗∗∗
[0.0114]
Two imes median sha e (m wo) -0.0593∗∗∗
[0.0163]
Low income high cos s (lihc) -0.101∗∗∗
[0.0176]
subjec i e -0.159∗∗∗
[0.0607]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 252040 252040 252040 86037
Fixed-e ec s eg ession. Dependen a iable: Cu en li e sa is ac ion (0=Low, 10=High). Clus-
e obus s anda d e o s in b acke s. All eg essions include he usual co a ia es. Da a: SOEP
(2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
Table A7: Fixed-e ec s Reg ession - heal h sa is ac ion
(1) (2) (3) (4)
Heal h sa is . Heal h sa is . Heal h sa is . Heal h sa is .
Ten pe cen ule ( p ) -0.0153
[0.0134]
Two imes median sha e (m wo) -0.0471∗∗
[0.0188]
Low income high cos s (lihc) -0.0335∗
[0.0200]
subjec i e -0.192∗∗∗
[0.0651]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
n ( o al) 250495 250495 250495 85928
Fixed-e ec s eg ession. Dependen a iable: Cu en heal h sa is ac ion (0= comple ely dissa is ied,
10=comple ely sa is ied). Clus e obus s anda d e o s in b acke s. All eg essions include he usual
co a ia es. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
37
Table A8: O de ed logis ic eg ession - equency o being happy
(1) (2) (3) (4)
F eq. happy F eq. happy F eq. happy F eq. happy
Ten pe cen ule ( p ) 0.908∗∗∗
[0.0185]
Two imes median sha e (m wo) 0.882∗∗∗
[0.0231]
Low income high cos s (lihc) 0.906∗∗∗
[0.0255]
subjec i e 0.837∗∗
[0.0736]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
Obse a ions ( o al) 227573 227573 227573 85873
Obse a ions (w. a ia ion in ou come) 180294 180294 180294 50033
Panel uni s (w. a ia ion in ou come) 28639 28639 28639 14803
Fixed-e ec s o de ed logi eg ession. Dependen a iable: F equency o being happy in he las 4 weeks
(1= e y seldom, 5= e y o en). Exponen ia ed coe icien s (odds a ios); Clus e obus s anda d e o s in
b acke s. All eg essions include he usual co a ia es. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
Table A9: O de ed logis ic eg ession - equency o being sad
(1) (2) (3) (4)
F eq. sad F eq. sad F eq. sad F eq. sad
Ten pe cen ule ( p ) 1.114∗∗∗
[0.0208]
Two imes median sha e (m wo) 1.123∗∗∗
[0.0283]
Low income high cos s (lihc) 1.098∗∗∗
[0.0291]
subjec i e 1.397∗∗∗
[0.122]
Socio-econ. con ols Yes Yes Yes Yes
Con ol o income po e y Yes Yes Yes Yes
Indi idual Fixed E ec s Yes Yes Yes Yes
Yea Fixed E ec s Yes Yes Yes Yes
Obse a ions ( o al) 227580 227580 227580 85873
Obse a ions (w. a ia ion in ou come) 198536 198536 198536 61117
Panel uni s (w. a ia ion in ou come) 32306 32306 32306 18162
Fixed-e ec s o de ed logi eg ession. Dependen a iable: F equency o being sad in he las 4
weeks (1= e y seldom, 5= e y o en). Exponen ia ed coe icien s (odds a ios); Clus e obus
s anda d e o s in b acke s. All eg essions include he usual co a ia es. Da a: SOEP (2022).
∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01
38
