1
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
www.na u e.com/scien i ic epo s
Using change ajec o ies o s udy
he impac s o mul i-annual habi a
loss on ledgling p oduc ion in an
old o es specialis bi d
E ic Le To o ec 1,2, Niina Käyhkö3, Ha i Hakka ainen2, Pe i Suo sa2, Esa Huh a4 & Samuli
Helle2
The loss and subdi ision o habi a in o smalle and mo e spa ially isola ed uni s due o human ac ions
has been shown o ad e sely a ec species wo ldwide. We examined how changes in old o es co e
du ing eigh yea s we e associa ed wi h he cumula i e numbe o ledged o sp ing a he end o s udy
pe iod in Eu asian eec eepe s (Ce hia amilia is) in Cen al Finland. We we e speci ically in e es ed in
whe he he ini ial le el o old o es co e mode a ed his ela ion. We applied a lexible and powe ul
app oach, la en g ow h cu e modelling in a s uc u al equa ion modeling (SEM) amewo k, o c ea e
ajec o ies desc ibing changes in old o es co e h ough ime, and s udied how his change a bo h
he e i o y co e and landscape scales impac ed ledging numbe s. Ou main inding was ha a he
e i o y co e scale he nega i e impac o habi a loss on ledging numbe s was lessened by he highe
le els o ini ial o es co e , while no associa ion was ound a he landscape scale. Ou s udy highligh s
a powe ul, bu cu en ly unde -u ilised me hodology among ecologis s ha can p o ide impo an
in o ma ion abou biological esponses o changes in he en i onmen , p o iding a mechanis ic way o
s udy how land co e dynamics can a ec species esponses.
The loss and subdi ision o habi a in o smalle and mo e spa ially isola ed uni s (i.e. habi a loss and agmen-
a ion) due o human ac ions ha e been shown o ad e sely a ec species wo ldwide. A as amoun o esea ch
s udying he e ec s o modi ied landscapes on species has been conduc ed using indi idual s a ic snapsho s o
habi a da a combined wi h biological da a1, 2. These s udies ha e g ea ly ad anced ou unde s anding o how
species espond o habi a s uc u e a di e en spa ial scales. Howe e , he empo al equency o biological da a
has o en been much highe han ha o habi a da a, i.e. ew landscape da a poin s in ela ion o he biological
da a poin s, which has limi ed he po en ial o make empo al in e ences o he impac s o landscape change on
species. Fo example, many s udies ha e been unable o model empo al pa e ns o habi a change a empo al
equencies ele an o he biological ques ion a hand, missing po en ially impo an dynamics3, o ha e been
unable o eliably dis inguish andom noise in ime se ies, caused by e.g. classi ica ion e o s, om ue change4.
As access o ee egional and global spa ial da a se s has subs an ially inc eased5, 6, ecen yea s ha e wi nessed
a g owing numbe o s udies quan i ying spa ially and empo ally explici habi a change in a ious ecosys ems7–9.
A numbe o empi ical s udies ha e also inco po a ed empo al aspec s in o s udies o habi a loss and agmen a-
ion. These s udies ha e shown clea nega i e e ec s o habi a loss and agmen a ion occu ing h ough ime on
o ganisms, being associa ed wi h dec eased numbe s o indi iduals and inc eased local ex inc ions10, dec eased
pe sis ence and occupancy o indi iduals wi hin landscapes11 as well as dec eased species ichness and abun-
dance12, 13. Howe e , a common sho coming sha ed by hese s udies is ha he empo al equency o habi a da a
has o en been spa se. Fo example, a s udy ha spans a ime pe iod ele an o he biological ques ion a hand
migh include only one o wo ime poin s, which migh lead o he s udy missing po en ial dynamics aking place
be ween he ime poin s. In addi ion, p e ious s udies ha e used he es ima es o change in subsequen sepa a e
1Depa men o Biological and En i onmen al Science, Uni e si y o Jy äskylä, P.O. Box 35, FI-40014, Jy äskylä,
Finland. 2Depa men o Biology, Uni e si y o Tu ku, FI-20014, Tu ku, Finland. 3Depa men o Geog aphy and
Geology, Uni e si y o Tu ku, FI-20014, Tu ku, Finland. 4Na u al Resou ces Ins i u e Finland, Ro aniemi Resea ch
Uni , P.O. Box 16, FI-96301, Ro aniemi, Finland. Ha i Hakka ainen is deceased. Co espondence and eques s o
ma e ials should be add essed o E.L. (email: [email p o ec ed])
Recei ed: 30 Sep embe 2016
Accep ed: 17 Ma ch 2017
Published: xx xx xxxx
OPEN
www.na u e.com/scien i ic epo s/
2
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
analyses o explo e species esponses, as opposed o conduc ing he en i e analysis igo ously in one s a is ical
amewo k aking he s a is ical unce ain y ela ed o change es ima ion app op ia ely in o accoun .
A mo e obus me hod o s udying en i onmen al changes is o use spa io- empo al change ajec o ies,
which can quan i y changes in en i onmen al a iables by conside ing mul iple yea s o da a simul aneously. This
makes hem a much mo e e ec i e me hod o cha ac e ising change han bi- empo al app oaches, whe e only
he di e ence be ween he beginning and end s a es a e conside ed14, 15. Compa ed o bi- empo al app oaches,
ajec o y-based me hods a e mo e eliable o quan i y a es and dynamics o changes16, less sensi i e o seasonal
and yea - o-yea a ia ion4 as well as o po en ial land co e classi ica ion e o s p esen in landscape da a. By
iden i ying changes in en i onmen al a iables h ough ime, ajec o y-based me hods lessen he impac o
ou lie s and missing da a poin s. A s a is ical amewo k combining change ajec o ies o en i onmen al da a
h ough ime wi h biological da a in a single s a is ical amewo k would hus be highly in o ma i e when linking
en i onmen al change o species esponses.
We s udied how he numbe o ledglings in he Eu asian eec eepe (Ce hia amilia is), he ea e he
eec eepe , in Cen al Finland was associa ed wi h changes in habi a co e in he immedia e icini y o he
nes box ( e i o y co e scale) and he su ounding landscape (landscape scale) o e a pe iod o eigh yea s.
We summed he numbe o ledged o sp ing pe nes box si e o e he en i e s udy pe iod, which enabled us
o quan i y he cumula i e impac s o con inuing habi a change on species ha could ha e been easily missed
when using s a ic snapsho s o habi a da a. We used pe cen co e o o es o e 50 yea s in age as an es ima e o
habi a amoun since i has p e iously been shown o be associa ed wi h e i o y occupancy17 and nes ling body
condi ion o eec eepe s18 a he e i o y co e scale, and nes p eda ion19, 20 a he landscape scale. Impo an ly,
we also in es iga ed whe he he le el o ini ial habi a co e mode a ed he in luence o change in habi a co e
on eec eepe ledging numbe s. The da a was analysed using la en g ow h cu e modelling in a s uc u al
equa ion modeling (SEM) amewo k21 (Fig.1), which is a lexible and powe ul mul i a ia e app oach, ye s ill
i ually unknown o ecologis s22. Fo bo h scales, sepa a ely, we decomposed he da a on annual amoun o
habi a co e in o ini ial le el o habi a a beginning o he s udy pe iod and in o change in habi a co e du ing
he s udy pe iod. These we e hen ela ed o he numbe o ledged o sp ing and he excess o ze o ledglings (a
ze o- in la ed model was used due o a p edominance o ze os), which enabled us o simul aneously model habi a
change and i s associa ion on he cumula i e numbe o ledglings pe nes box si e in a single s a is ical model.
Based on ou p e ious esul s on his sys em17, 20, we hypo hesised ha : (1) dec easing habi a co e a he
e i o y co e scale would dec ease he numbe o ledged o sp ing by educing he o al numbe o nes ing
e en s, (2) dec easing habi a co e a he landscape scale would inc ease he numbe o ledged o sp ing because
o lowe p eda ion p essu e, and (3) a high le el o ini ial habi a co e in he nes ing si e would p o ec he si es
agains he nega i e in luence o dec easing habi a co e .
Figu e 1. G aphical ep esen a ion o baseline la en g ow h cu e models used o examine he in luence o
change in habi a co e , ini ial habi a co e and hei in e ac ion on he cumula i e numbe o ledged o sp ing
a he end o he s udy pe iod. The boxes on he le ep esen obse ed habi a da a collec ed om indi idual
yea s om which ini ial habi a co e and change in habi a co e , as well as hei in e ac ion a e de i ed om.
Single-headed a ows in he la en g ow h cu e pa o he model (on he le side) ep esen ac o loadings,
connec ing he obse ed annual habi a co e and unobse ed la en in e cep and slope ac o s. Numbe s along
hese a ows mean ixed alues, while as e isks deno e es ima ed alues om he da a. Single-headed a ows
in he s uc u al pa o he model (on he igh side) ep esen s uc u al pa h coe icien s examining how
in e cep and slope ac o s and hei in e ac ion in luence cumula i e ledgling numbe . The a ows o igina ing
om he connec ing do be ween ini ial habi a co e and change in habi a co e ep esen la en in e ac ions
be ween he la en ac o s. Co a iance be ween ini ial habi a co e and change in habi a co e is depic ed as
a double-headed a ow. Sho a ows poin ing a he habi a a iables ep esen hei esidual a iances. Please
no e ha hese baseline models excluded esidual co a iances be ween habi a co e ha we included in inal
models.
www.na u e.com/scien i ic epo s/
3
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
Resul s
A he beginning o he s udy pe iod, nes box si es had an a e age (SD) o 66.8% (20.9) and 47.7% (14.8) o habi-
a co e a he e i o y co e and landscape scales, espec i ely (Fig.2). The pa o he la en g ow h cu e model
desc ibing change in habi a co e in hese nes box si es indica ed ha , as expec ed om he s udy design, ini ial
habi a co e a ied signi ican ly be ween he nes box si es a bo h scales. This was e iden by he signi ican a i-
ances o he co esponding la en in e cep s a bo h scales (Table1). The e was also a s a is ically signi ican a e -
age dec ease in habi a co e a bo h spa ial scales. Habi a was los on a e age a he a e o 2.2 and 1.4% pe yea
a he e i o y co e and landscape scales, espec i ely (Table1, Fig.2). We also ound a ia ion om his a e age
change be ween nes box si es, sugges ing ha he a e o habi a loss a ied be ween nes box si es (al hough a
he landscape scale he signi icance o a iance o change in habi a co e was ma ginal; Table1). The co a iance
be ween he ini ial le el o habi a co e and change in habi a co e was signi ican o bo h scales: a he e i o y
co e scale, he nes box si es wi h high le els o ini ial habi a co e los less habi a han hose wi h less ini ial
habi a , while his ela ion was in opposi e di ec ion a he landscape scale (Table1).
When associa ing changes in habi a co e wi h he cumula i e numbe o eec eepe ledglings a he end o
he s udy pe iod, we ound ha a he e i o y co e scale, ini ial habi a co e , change in habi a co e and hei
in e ac ion did no p edic he excess o ze o ledglings (si es ha p oduced no ledged o sp ing) (ze o-in la ed
logis ic model in Table1, Fig.1). Ins ead, we ound a s a is ically signi ican in e ac ion be ween change in habi a
co e and ini ial habi a co e p edic ing he numbe o ledglings (nega i e binomial model in Table1). Tha is,
inc easing habi a loss du ing he s udy pe iod, which dec eased he cumula i e numbe o ledglings, was a en-
ua ed in nes box si es wi h high ini ial habi a co e . When conside ing a iance-s anda dised esul s, in nes
box si es wi h ini ial habi a co e one s anda d de ia ion abo e he mean, a dec ease o one s anda d de ia ion
in habi a change (i.e. mo e habi a loss) dec eased he expec ed numbe o ledglings by 4.3% (Fig.3). In nes
box si es wi h ini ial habi a co e one s anda d de ia ion below he mean, a one s anda d de ia ion dec ease in
habi a change dec eased he expec ed numbe o ledglings by 32.3% (Fig.3). Ini ial habi a co e , change in
habi a co e o hei in e ac ion we e no associa ed wi h he excess o ze o ledglings no he numbe o ledged
o sp ing a he landscape scale (Table1, Fig.4).
Discussion
A numbe o p e ious s udies ha e shown ha habi a loss, quan i ied om snapsho s o habi a da a, in luence
impo an ai s associa ed wi h indi idual ep oduc i e success, such as nes si e occupancy, clu ch size and he
numbe o ledged o sp ing17, 23, 24. This s udy complemen s his li e a u e by explici ly associa ing mul i-annual
changes in habi a co e wi h nes - si e le el p oduc ion o o sp ing in a single s a is ical amewo k. The
eec eepe nes ing si es in ou s udy a ea expe ienced a clea dec ease in he co e o old o es , which is he
main habi a o hese bi ds. On a e age, nes box si es los oughly 17% habi a co e a bo h spa ial scales du ing
ou s udy pe iod o eigh yea s. Taking ad an age o a la en g ow h cu e modelling in a s uc u al equa ion
modeling amewo k we ound ha a he e i o y co e scale he associa ion be ween a e o habi a loss and he
Figu e 2. Box plo s showing change in habi a co e a he e i o y co e (100 m) (a) and landscape (600 m)
(b) scales du ing he s udy pe iod o eigh yea s. The box plo s show aw da a o pe cen co e o habi a (old
o es ) o nes box si es o each yea sepa a ely. The ops and bo oms o he g ey boxes ep esen he 75 h and
25 h pe cen iles, espec i ely, while he lines in he middle o he g ey boxes ep esen he median and he whi e
diamonds show he mean. The whiske s abo e and below he g ey boxes ep esen 1.5 imes he in e -qua ile
ange, and he indi idual poin s ep esen alues beyond hese.
www.na u e.com/scien i ic epo s/
4
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
numbe o ledged eec eepe young seemed o be mode a ed by he amoun o ini ial habi a co e : he ha m ul
in luence o habi a loss on ledgling numbe s was educed in nes box si es wi h high ini ial habi a co e com-
pa ed wi h hose wi h less ini ial habi a co e . A he landscape scale, we did no ind any in luence o empo al
habi a loss on ledging numbe s.
We ound suppo o he p o ec i e e ec o high le els o ini ial habi a co e agains he nega i e impac
o empo al habi a loss on ledging numbe s is eec eepe s. This e ec seemed a he s ong as in nes box si es
wi h ini ial habi a co e one s anda d de ia ion below he mean, a one s anda d de ia ion dec ease in habi a
co e o e ime dec eased he expec ed numbe o ledglings by 32.3%. This inding highligh s he impo ance
o conside ing mul iple habi a cha ac e is ics simul aneously in s udies o habi a - ela ed in luences on species
esponses. P e ious s udies in ou s udy popula ion, using s a ic snapsho s o habi a s uc u e, ha e also shown
a nega i e in luence o habi a loss on ine- uned esponses such as physiological s ess in nes lings18, 19. Ou
app oach o cha ac e ising changes in habi a co e combined wi h agg ega ed biological da a enabled us o quan-
i y sub le in luences o con inuing habi a change on species ha could ha e been easily missed when using s a ic
100 m 600 m
Es ima e SE P- alue Es ima e SE P- alue
S uc u al pa h coe icien s
Nega i e binomial coun model
Ini ial habi a co e 0.02 0.07 0.776 −0.011 0.079 0.895
Change in habi a co e 2.886 1.31 0.060 0.536 2.326 0.818
Ini ial co e × Change −0.361 0.195 0.064 0.234 0.543 0.666
Ze o-in la ed logis ic model
Ini ial habi a co e −1.75 1.187 0.140 −0.301 0.391 0.441
Change in habi a co e 33.558 28.205 0.234 −4.755 11.152 0.818
Ini ial co e × Change −4.21 3.797 0.268 1.437 2.688 0.593
La en a iable means
Ini ial habi a co e 6.775 0.149 <0.001 4.806 0.103 <0.0001
Change in habi a co e −0.22 0.022 <0.001 −0.142 0.008 <0.0001
La en a iable a iances
Ini ial habi a co e 3.205 0.491 <0.001*1.998 0.184 <0.0001*
Change in habi a co e 0.033 0.014 0.001*0.005 0.003 0.001*
Basis coe icien s
1999 0 Fixed — 0 Fixed —
2001 0.762 0.438 0.082 1.067 0.237 <0.001
2002 2.835 0.605 <0.001 1.832 0.249 <0.001
2003 2.834 0.667 <0.001 0.503 0.275 0.068
2005 4.315 0.488 <0.001 3.029 0.231 <0.001
2006 7 Fixed — 7 Fixed —
La en a iable co a iances
Ini ial habi a co e –
Change in habi a co e 0.121 0.038 0.001*−0.033 0.011 <0.001*
Residual co a iances
Yea s 1999, 2001 0.68 0.241 0.005 0.048 0.022 0.029
Yea s 2001, 2002 0.387 0.143 0.007 0.016 0.01 0.122
Yea s 2002, 2003 0.698 0.214 0.001 0.014 0.011 0.180
Yea s 2003, 2005 0.339 0.121 0.005 0.054 0.015 <0.0001
Yea s 2005, 2006 0.59 0.24 0.014 0.041 0.039 0.295
Residual a iances
Yea 1999 1.37 0.187 <0.0001 0.242 0.047 <0.0001
Yea 2001 1.37 0.187 <0.0001 0.092 0.018 <0.0001
Yea 2002 1.37 0.187 <0.0001 0.103 0.02 <0.0001
Yea 2003 1.37 0.187 <0.0001 0.07 0.02 0.001
Yea 2005 1.37 0.187 <0.0001 0.124 0.02 <0.0001
Yea 2006 1.37 0.187 <0.0001 0.144 0.098 0.141
Table 1. Resul s o wo sepa a e la en g ow h cu e models examining how change in habi a co e , ini ial
le el o habi a co e and hei in e ac ion a he e i o y co e (100 m) and he landscape (600 m) scales we e
associa ed wi h he o al numbe o ledged o sp ing p oduced du ing he s udy pe iod. No e ha eg ession
coe icien s on nega i e binomial and ze o-in la ed logis ic models a e on log and logi scales, espec i ely, and
ha he p- alues o ac o a iances a e hal ed because, by de ini ion, hese canno be nega i e (Hox 2010).
*S a is ical signi icance de e mined wi h likelihood a io es .
www.na u e.com/scien i ic epo s/
5
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
snapsho s o habi a da a. The in luence o change in habi a co e on ledging numbe s a he e i o y co e scale
is in line wi h he esul s o Suo sa e al.17, whe e he co e o old o es immedia ely su ounding he nes si e had
a s ong posi i e associa ion wi h e i o y occupancy. In a p e ious s udy we did no ind any e idence ha high
quali y indi iduals occupied he la ges o es pa ches25, sugges ing ha he esul s seen he e a e no likely due o
he highes quali y indi iduals b eeding in si es wi h he mos old o es . In e es ingly, we did no ind e idence
ha changes in habi a co e o ini ial habi a co e a he landscape scale had any impac on ledging numbe s.
This sugges s ha he main mechanism impac ing he long- e m numbe o ledglings in his species was nes
occupancy and esou ce a ailabili y, which a e mainly de e mined by habi a a iables su ounding he nes 17,
a he han nes p eda ion, which is in luenced by habi a a iables a he landscape scale19.
Despi e he obus s a is ical me hods used in his s udy, he e a e some po en ial sou ces o e o ha may
ha e in luenced ou esul s. Fi s , i is possible ha he de ini ion o habi a used in his s udy was oo b oad o
show he ull in luence o change in habi a co e on eec eepe s. Suo sa e al.17 showed clea a ea-sensi i i y in
he occupancy o eec eepe nes box si es, bu he wood olume used in ha s udy was o e 150 m3/ha, which
was highe han he le el used in his s udy (100 m3/ha). Howe e , se e al p e ious s udies ha e shown ha he
same habi a de ini ion used in his s udy can a ec bo h physiological and li e his o y ai s in eec eepe s18–20.
Second, e o s in he classi ica ion o sa elli e images migh ha e esul ed in alse change in he esul s, o exam-
ple showing dec eases in habi a co e in si es ha had s able habi a co e . Howe e , a p e ious classi ica ion
accu acy assessmen 20 o he la es image used in his s udy showed ha classi ica ion e o s should no ha e
been a se ious p oblem he e. In addi ion, ou me hod o i ing a end h ough he ime se ies o each nes box
si e lessened he impac o po en ial classi ica ion e o s. Thi d, we did no ake he po en ial impac o wea he
on eec eepe ledging numbe s in o accoun due o sample size limi a ions in he s a is ical models. Al hough
wea he condi ions du ing he nes ing season can ha e s ong impac s on ep oduc i e success in bi ds26 we did
no ind any impac o empe a u e o ain du ing he nes ing season on eec eepe ep oduc i e success in a
p e ious s udy20, sugges ing ha i s omission unlikely p oduced se ious bias o ou esul s. Finally, owing o ou
mode a e sample size (i.e. he numbe o nes box si es), ou analysis migh ha e su e ed om inadequa e s a-
is ical powe o de ec sub le e ec s o habi a loss on eec eepe ledging numbe s. Gene al ules ega ding he
sample sizes needed a e ha d o come by in la en g ow h cu e li e a u e and depend e.g. on model complexi y21.
In addi ion, a lack o a p io i knowledge o he expec ed e ec s sizes means ha pos hoc powe analyses a e no
sui able.
Ou s udy shows he u ili y o a powe ul modelling app oach by demons a ing he applicabili y o la en
g ow h cu e modelling in simul aneously desc ibing en i onmen al change and ela ing his o an aspec o
ep oduc i e success. La en g ow h cu e modelling is akin o mul ile el (o hie a chical o mixed) model-
ling ha mos ecologis s a e amilia wi h. Howe e , i p o ides a mo e lexible amewo k o model complex
Figu e 3. G aphical ep esen a ion o he mode a ing in luence o he ini ial le el o habi a co e in a nes box
si e on he associa ion be ween change in habi a co e and he o al numbe o ledged o sp ing a he end o
he s udy pe iod a he e i o y co e scale. The solid line ep esen s he esponse, in s anda d de ia ions, o he
sum o ledged o sp ing in esponse o a change in habi a co e . The do ed line ep esen s he same associa ion
o nes box si es wi h ini ial habi a co e one s anda d de ia ion abo e he a e age, while he dashed line
ep esen s nes box si es wi h ini ial habi a co e one s anda d de ia ion below he mean. In nes box si es
whe e ini ial habi a co e was below he mean le el, eec eepe ledging numbe was mo e s ongly impac ed
by habi a change han in si es whe e ini ial habi a co e was a o abo e he mean le el.

www.na u e.com/scien i ic epo s/
6
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
mul i a ia e longi udinal p ocesses han a mul ile el amewo k when he numbe o le els is limi ed o h ee
o less21. In addi ion, he in e ac ion be ween ini ial habi a co e and subsequen change in habi a co e on
eec eepe ledging numbe s was s aigh o wa d o model in a SEM amewo k, bu would ha e been cum-
be some o do in s anda d mul ile el amewo k27. Ins ead o s udying he e how changes in habi a co e in lu-
ence ledging numbe s o he species we could ha e also included ime- a ian and ime–in a ian p edic o s
(e.g. clima ic o policy decisions) di ec ly in luencing en i onmen al change pe se as well as species ou come(s).
Inclusion o measu emen e o in a iables o in e es by using common ac o s is also eadily implemen ed in
a la en g ow h cu e app oach, which may esul in less causally inconsis en es ima es and inc eased s a is ical
powe 28. As a downside, complex models need gene ally mo e da a, which may hinde SEM’s applicabili y in
small da a se s. We howe e expec his o be no obs acle in he upcoming e a o big da a.
Since oughly 85% o he o es ed a ea in Finland is no p o ec ed29, he ecologically sus ainable managemen
o o es s subjec ed o comme cial ac i i ies is c i ical o he o e all e en ion o biodi e si y in he coun y. As
such, de e mining he e ec s o a es o change on biological esponses p o ides impo an in o ma ion ha
can be used o plan long- e m managemen o o es s ands. In e ms o o es managemen , he esul s o his
s udy suppo he c ea ion o ewe la ge a eas o habi a ha can p o ec species agains changes in habi a co e ,
e sus p ese ing mul iple small pa ches whe e he nega i e e ec s o habi a loss will be magni ied, leading o
local ex inc ions and smalle popula ion capaci y30. Howe e , i is impo an o keep in mind ha his s udy was
conduc ed on a single species, and as such hese esul s migh be species-speci ic. We sugges ha s udies using
a simila change de ec ion me hodology bu inspec ing popula ion- and communi y- le el esponses in di e en
landscapes could yield managemen ecommenda ions ha a e applicable in b oade con ex s. In addi ion, he
me hodology ou lined he e o e s a way o inspec po en ial h eshold31 a es o habi a change a e which i s
e ec s clea ly inc ease.
Figu e 4. Sca e plo s showing he associa ion be ween he le el o ini ial habi a co e and he numbe o
ledged o sp ing summed pe nes box si e, and he associa ion be ween change in habi a co e , measu ed by
ac o sco es, and he summed numbe o ledged o sp ing. Panel (a) shows he associa ion be ween he ini ial
le el o habi a co e and he numbe o ledged o sp ing a he e i o y co e scale, while panel (b) shows he
associa ion be ween change in habi a co e and ledged o sp ing a he same spa ial scale. Panels (c) and (d)
show he same associa ions a he landscape scale. No e ha ac o sco es a e se on an a bi a y scale, wi h he
mean alue co esponding o he alue o he ac o mean (see Table1).
www.na u e.com/scien i ic epo s/
7
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
Ma e ials and Me hods
T eec eepe da a. The eec eepe is a small a ea-sensi i e passe ine, which p e e s old o es s (>50 yea s
old) as b eeding habi a 17. I eeds by sea ching o in e eb a es on ee unks32 and cons uc s a nes unde a lap
o loose ba k o in c e ices in ee unks33, bu also eadily accep specially designed nes boxes. The da a used in
his s udy we e collec ed be ween 1999 and 2006 om a s udy si e co e ing 1150 km2 in Cen al Finland (cen ed
on 62°37′N, 26°20′E). Du ing he s udy pe iod his a ea was subjec o in ensi e comme cial o es y whe eby
o es s a e clea cu , eplan ed and hinned in a oughly 80-yea cycle.
The s udy a ea consis ed o a o al o 241 nes box si es selec ed o include agmen ed, in e media e and
un agmen ed landscapes. Each si e con ained wo iden ically sized nes boxes placed 30 m apa o acili a e
po en ial second b eeding a emp s wi hin he b eeding season. Al hough some indi idual eec eepe s nes ed
in na u al ca i ies in ou s udy si e, hese numbe s we e so low (0–3 pe yea o he en i e s udy si e) ha hei
impac on he s udy mus ha e been negligible. Th oughou he b eeding season (Ap il – July) each nes box si e
was isi ed a leas wo imes o iden i y occupied si es. Occupied si es we e u he isi ed h oughou he b eed-
ing season o de e mine i s and second b eeding a emp s, coun he numbe o eggs laid, ing and measu e nes -
lings and cap u e and measu e he pa en s. P eda ion e en s could be iden i ied on he s a e o he nes , and he
inal numbe o ledged o sp ing was de e mined by sub ac ing he numbe o deceased chicks om he numbe
o nes lings. Mo e de ailed in o ma ion o he s udy p o ocols can be ound in Huh a e al.19. Fo he cu en s udy,
he numbe o ledglings was no summed pe indi idual emale eec eepe because hese only nes ed on a e age
(SD) 1.3 (0.62) imes in ou s udy popula ion. Ins ead, he numbe o ledglings was summed pe each nes box
si e o he whole s udy pe iod. This was done o maximise he numbe o nes box si es included in he s udy.
Despi e he high u no e o indi iduals b eeding in a single nes box si e we did no ind any e idence ha mo e
expe ienced indi iduals o hose in be e condi ion occupied si es wi h he mos old o es 20. In o de o assess
change in habi a co e we used only nes box si es om which a leas wo consecu i e yea s o landscape da a
was a ailable om he same loca ion (i.e. he nes box si e had no been mo ed), which esul ed in 213 nes box
si es. These nes box si es had an a e age (SD) o 5.3 (0.96) yea s o habi a da a (ou o six sepa a e habi a da a
poin s). F om he nes ing da a, we excluded nes ing e en s whe e expe imen al p ocedu es o p e ious s udies
had aken place du ing he s udy du a ion (n = 149), which esul ed in a o al o 1009 sepa a e nes ing e en s by
475 indi idual emale eec eepe s. Each nes ing si e p oduced an a e age o 14.4 (10.6) ledged o sp ing du ing
he en i e s udy pe iod. Nes box si es had an a e age o 7.78 (0.57) yea s o nes ing da a: wo si es had ou yea s
o da a, ou si es had six yea s, 30 si es had se en yea s, and 177 had no missing yea s.
All me hods we e ca ied ou in acco dance wi h ele an guidelines and egula ions, and ou s udy complies
wi h he cu en laws o Finland. All expe imen al p o ocols we e app o ed by he En i onmen al Cen e o
Cen al Finland.
Quan i ying changes in habi a co e . Habi a change da a we e gene a ed om Landsa 5 Thema ic
Mappe sa elli e images ( esolu ion 30 m × 30 m) o he yea s 1999, 2001, 2002, 2003, 2005 and 2006 (2000 and
2004 we e missing due o lack o cloud- ee images), downloaded om he Uni ed S a es Geological Su ey
Global Visualiza ion Viewe se ice (h p://glo is.usgs.go ). The en i e s udy a ea was co e ed wi h one image,
and o each yea we selec ed one image wi h he minimum amoun o cloud co e , aken be ween he i s o
May and las o Sep embe . Fo 2006 a cloud- ee composi e o wo sequen ial cloudy images was used. Each sa -
elli e image was classi ied in o wo classes wi h supe ised classi ica ion: based on p e ious s udies17, 18, old o es
(wood olume o e 100 m3/ha, ci ca 50 yea s o olde ) was classi ied as habi a o eec eepe s, while he ma ix
class con ained all o he land classes (e.g. buil -up a eas, wa e , ields and younge o es classes). P ocessing and
classi ica ion o sa elli e images is explained in mo e de ail in Supplemen a y ma e ial 1.
Habi a co e was calcula ed a he e i o y co e and landscape scales o s udy po en ial scale-dependen di -
e ences o he e ec s o change in habi a co e . A ci cula a ea ex ending 100 m om he nes box pai , co e ing
an a ea o abou 3ha, was used o desc ibe he e i o y su ounding he nes box si e. This was conside ed as he
smalles possible a ea om which habi a a ea could be quan i ied, conside ing he pixel size o 30 m × 30 m o
he o iginal sa elli e images. Al hough his was smalle han he adius o 200 m used in p e ious s udies19, 20 he e
is e idence ha inc eased habi a co e in he immedia e icini y o he nes box si e has a posi i e in luence on
a key pa ame e o habi a sui abili y, he p obabili y o nes si e occupancy17. In addi ion, his coincides oughly
wi h he dis ance o leas 70 m ha male eec eepe s de end a ound he nes 33. A ci cula a ea ex ending 600 m,
co e ing abou 113ha, ep esen ed he landscape scale. The landscape scale was also included since he co e o
habi a a his spa ial scale has been shown o ha e bo h nega i e and posi i e impac s on nes p eda ion p obabil-
i y in ou s udy sys em19, 20. Pixels ha had a leas 50% o e lap wi h he ci cula bu e s ep esen ing bo h spa ial
scales we e used o assessing o es co e a ound he nes box si e. Fo each nes box si e and yea , he co e o
old o es was calcula ed o bo h scales using F ags a s 3.434. To a oid inconsis encies in habi a da a be ween
clouded and cloud- ee e i o ies, he pe cen age co e o habi a was used ins ead o absolu e a ea. We hen used
a change ajec o y-based app oach o quan i y he change in habi a co e h ough he s udy pe iod o each nes
box si e sepa a ely. We also quan i ied he ini ial le el o habi a co e o each indi idual nes box si e, o explo e
whe he he ini ial le el o habi a co e a he s a o he habi a se ies had a mode a ing e ec on he e ec o
change in habi a co e .
S a is ical me hods. The di ec and mode a ed in luence o ini ial co e and change in habi a co e on he
cumula i e numbe o ledglings o eec eepe s was quan i ied using la en g ow h cu e model in a s uc u al
equa ion modeling (SEM) amewo k21, 35. La en g ow h cu e modelling is an applica ion o SEM o longi udinal
da a analysis, designed o pa i ion a ia ion o po en ially mul iple p ocesses in o wi hin- and be ween-subjec
a ia ion36 and, hus, i sha es many cha ac e is ics wi h mul ile el modeling21. La en g ow h cu e models used
he e desc ibe ini ial habi a co e , a e o change o e ime and hei co a iance as la en (unobse ed) in e cep
www.na u e.com/scien i ic epo s/
8
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
and slope ac o s wi h andom coe icien s36. He e, he in e cep ac o ep esen s he ini ial co e o habi a a he
beginning o he s udy pe iod and he slope ac o ep esen s he change in habi a co e du ing he s udy pe iod.
The means o hese pa ame e s desc ibe he a e age empo al habi a co e me ics be ween nes box si es (i.e.
co espond o ixed pa ame e s), while hei a iances desc ibe he be ween-nes box si e a ia ion in hese me -
ics (i.e. co espond o andom pa ame e s). In la en g ow h cu e models, he pa hs (i.e. loadings) om la en
in e cep ac o o he obse ed ou comes a e ixed o 1 while he loadings o la en slope ac o (i.e. ime sco es)
ep esen he measu emen o ime since he measu emen o ini ial habi a co e and hey de e mine he o m
and cen ing (baseline) poin o change ajec o y36. The o m o a e o change in la en g ow h cu e modelling
can ake many shapes (e.g. linea , quad a ic o exponen ial)36. He e, we allowed he ime sco es (excep he i s
and he las ime sco es ha we e used o se he ajec o y scale o ob ain a change pe yea o e ou s udy pe iod)
o be eely es ima ed om he da a (i.e. we used basis unc ions), because p elimina y inspec ion o he da a
sugges ed nonlinea change o no clea pa ame ic shape36 (Fig.2).
In la en g ow h cu e modelling i is possible o examine associa ions be ween mul iple longi udinal p o-
cesses in he same model36. Howe e , owing o ou modes numbe o nes box si es (n = 213), he inclusion
o habi a change a bo h scales in addi ion o he longi udinal change in annual eec eepe ledgling numbe s
du ing he s udy pe iod would ha e esul ed in a g ossly o e i ed model (47 es ima ed pa ame e s a minimum,
depending on he model complexi y). The e o e, he in luence o change in habi a co e on he ledgling num-
be s o eec eepe s was analysed sepa a ely o he e i o y co e and landscape scales and by using he sum o
he numbe o ledglings o e he s udy yea s (i.e., he cumula i e numbe o ledglings was modelled as a dis al
ou come). The p opo ion o habi a co e a bo h scales we e ea ed as con inuous esponse a iables and, o aid
model con e gence due o oo small o oo la ge ac o a iances, habi a co e alues we e di ided by 10 p io
o he analyses.
Since he cumula i e numbe o ledglings pe nes ing si e a he end o he s udy pe iod was a coun a iable,
we s a ed by selec ing he app op ia e wo king e o dis ibu ion o his a iable. This was accomplished by
using sample-size adjus ed Bayesian in o ma ion c i e ion (SABIC), which is shown o pe o m well in model
selec ion asks in a SEM amewo k37, 38. Owing o he p edominance o ze oes in he cumula i e numbe o
ledglings (Supplemen a y Fig.S2), we con as ed he ollowing po en ial e o dis ibu ions: Poisson and nega-
i e binomial dis ibu ions as well as hei ze o-in la ed coun e pa s ha conside a mix u e o esponse dis i-
bu ions39. A bo h scales, ze o-in la ed nega i e binomial dis ibu ion clea ly i ed he da a bes (Supplemen a y
TableS2.1). This mean ha we now had wo eg essions o es ima e wi h espec o ledgling numbe : one ha
models how habi a co e a ibu es p edic ze o o highe numbe o ledglings (i.e. he coun pa o he mix-
u e) and ano he on how he same habi a a ibu es p edic ze o ledglings only (i.e. he ze o-in la ion pa o he
mix u e using logis ic eg ession). Nex , using he same model selec ion app oach, we compa ed di e en esidual
a iance s uc u es o annual habi a co e because a miss-speci ied esidual s uc u e can bias he a iance es i-
ma e o in e cep and slope ac o s, as well as hei co a iance40. This was done by compa ing models assuming
he ollowing esidual a iance s uc u es in he ollowing o de : homogeneous esiduals, he e ogeneous esidu-
als, homogeneous esiduals wi h co a iances wi h adjacen ime poin s and he e ogeneous esiduals wi h co a i-
ances wi h adjacen ime poin s. Co a iances among a iances be ween adjacen ime poin s con ol o po en ial
empo al au oco ela ion be ween adjacen ime poin s no accoun ed by he in e cep and slope ac o s21. Fo he
e i o y co e scale, he bes model included homogenous e o s and es ima ed hei esidual co a iances. Fo he
landscape le el, he bes model i was ob ained by es ima ing he e ogeneous esiduals and es ima ing he esidual
co a iances be ween adjacen annual habi a co e measu emen s (Supplemen a y TableS2.2).
We used a obus maximum likelihood (MLR) es ima o whe e missing da a assumed o be missing a an-
dom was handled using ull in o ma ion maximum likelihood es ima ion41 using Mplus 7.342. In ull in o ma ion
maximum likelihood es ima ion, models a e es ima ed by gi ing he obse a ions ha ing mo e da a poin s mo e
weigh compa ed o obse a ions ha ing less da a poin s. This app oach allows missing alues o dependen
a iables, meaning he e bo h he eco ds o annual habi a co e s as well as he numbe s o ledglings. The in lu-
ence o he in e ac ion be ween ini ial habi a co e and change in habi a co e on he numbe o ledglings was
es ima ed using he la en mode a ed s uc u al equa ions me hod43. No commonly used es s and absolu e i
indexes a e a ailable o assess model i o he da a, because he es ima ion o he cu en model equi es he usage
o indi idual da a and hence a iable(s) means, a iances and co a iances a e no su icien o model es ima ion.
We es ed he signi icance o a iance componen s, as well as hei co a iance, o he slope (change in habi a
co e ) and in e cep (ini ial habi a co e ) using likelihood a io es s27. We we e no able o es o spa ial au o-
co ela ion since SEM does no p oduce adi ional esiduals om which spa ial au oco ela ion can be es ed o .
Howe e , he esul s o ou p e ious s udy20 showed ha he esiduals om models inspec ing he in luence o
habi a agmen a ion on he numbe o ledglings did no display spa ial au oco ela ion. The e o e, e en hough
he e was some o e lap be ween nes box si es a he landscape (600 m) scale pseudo eplica ion should no be an
issue he e.
Re e ences
1. Fah ig, L. E ec s o habi a agmen a ion on biodi e si y. Annu. Re . Ecol. E ol. Sys . 34, 487–515, doi:10.1146/annu e .
ecolsys.34.011802.132419 (2003).
2. Tu ne , M. G. Landscape Ecology: Wha Is he S a e o he Science? Annu. Re . Ecol. E ol. Sys . 36, 319–344, doi:10.1146/annu e .
ecolsys.36.102003.152614 (2005).
3. Lu, D., Mausel, P., B ondízio, E. & Mo an, E. Change de ec ion echniques. In . J. Remo e Sens. 25, 2365–2401,
doi:10.1080/0143116031000139863 (2004).
4. Lambin, E. F. Change de ec ion a mul iple empo al scales: seasonal and annual a ia ions in landscape a iables. Pho og amm. Eng.
Remo e Sens. 62, 931–938 (1996).
5. Vogelmann, J. E., Tolk, B. & Zhu, Z. Moni o ing o es changes in he sou hwes e n Uni ed S a es using mul i empo al Landsa da a.
Remo e Sens. En i on. 113, 1739–1748, doi:10.1016/j. se.2009.04.014 (2009).
www.na u e.com/scien i ic epo s/
9
Scien i ic RepoR s | 7: 1874 | DOI:10.1038/s41598-017-02072-w
6. Sex on, J. O., U ban, D. L., Donohue, M. J. & Song, C. Long- e m land co e dynamics by mul i- empo al classi ica ion ac oss he
Landsa -5 eco d. Remo e Sens. En i on. 128, 246–258, doi:10.1016/j. se.2012.10.010 (2013).
7. Sade , S. A., Hayes, D. J., Hepins all, J. A., Coan, M. & Soza, C. Fo es change moni o ing o a emo e biosphe e ese e. In . J. Remo e
Sens. 22, 1937–1950, doi:10.1080/01431160117141 (2001).
8. Röde , A., Udelho en, T., Hill, J., del Ba io, G. & Tsiou lis, G. T end analysis o Landsa -TM and -ETM+ image y o moni o
g azing impac in a angeland ecosys em in No he n G eece. Remo e Sens. En i on. 112, 2863–2875, doi:10.1016/j. se.2008.01.018
(2008).
9. F ase , R. H., Ol ho , I., Ca iè e, M., Deschamps, A. & Poulio , D. De ec ing long- e m changes o ege a ion in no he n Canada
using he Landsa sa elli e image a chi e. En i on. Res. Le . 6, 45502, doi:10.1088/1748-9326/6/4/045502 (2011).
10. Boulinie , T. e al. Fo es agmen a ion and bi d communi y dynamics: in e ence a egional scales. Ecology 82, 1159–1169,
doi:10.1890/0012-9658(2001)082[1159:FFABCD]2.0.CO;2 (2001).
11. Pen e iani, V. & Fai e, B. E ec s o ha es ing imbe s ands on goshawk nes ing in wo Eu opean a eas. Biol. Conse . 101,
211–216, doi:10.1016/S0006-3207(01)00068-4 (2001).
12. Knick, S. T. & Ro enbe y, J. T. Ghos s o Habi a s pas : Con ibu ion o Landscape Change o Cu en Habi a s Used by Sh ubland
Bi ds. Ecology 81, 220–227, doi:10.1890/0012-9658(2000)081[0220:GOHPCO]2.0.CO;2 (2000).
13. Ri enhouse, C. D. e al. Land-Co e Change and A ian Di e si y in he Con e minous Uni ed S a es: Land-Co e Change and
A ian Di e si y. Conse . Biol. 26, 821–829, doi:10.1111/j.1523-1739.2012.01867.x (2012).
14. Coppin, P., Jonckhee e, I., Nackae s, K., Muys, B. & Lambin, E. Digi al change de ec ion me hods in ecosys em moni o ing: a
e iew. In . J. Remo e Sens. 25, 1565–1596, doi:10.1080/0143116031000101675 (2004).
15. Kennedy, R. E., Cohen, W. B. & Sch oede , T. A. T ajec o y-based change de ec ion o au oma ed cha ac e iza ion o o es
dis u bance dynamics. Remo e Sens. En i on. 110, 370–386, doi:10.1016/j. se.2007.03.010 (2007).
16. Gillande s, S. N., Coops, N. C., Wulde , M. A., Ge gel, S. E. & Nelson, T. Mul i empo al emo e sensing o landscape dynamics and
pa e n change: desc ibing na u al and an h opogenic ends. P og. Phys. Geog . 32, 503–528, doi:10.1177/0309133308098363
(2008).
17. Su o sa , P. e al. Th esholds in selec ion o b eeding habi a by he Eu asian eec eepe (Ce hia amilia is). Biol. Conse . 121,
443–452, doi:10.1016/j.biocon.2004.05.014 (2005).
18. Suo sa, P. e al. Fo es managemen is associa ed wi h physiological s ess in an old–g ow h o es passe ine. P oc. R. Soc. Lond. B
Biol. Sci. 270, 963–969, doi:10.1098/ spb.2002.2326 (2003).
19. Huh a, E. e al. Fo es agmen a ion inc eases nes p eda ion in he Eu asian eec eepe . Conse . Biol. 18, 148–155, doi:10.1111/
cbi.2004.18.issue-1 (2004).
20. Le To o ec, E. e al. Habi a agmen a ion and ep oduc i e success: a s uc u al equa ion modelling app oach. J. Anim. Ecol. 82,
1087–1097, doi:10.1111/1365-2656.12075 (2013).
21. Li le, T. D. Longi udinal s uc u al equa ion modeling (Guil o d P ess, 2013).
22. G ace, J. B. S uc u al equa ion modeling and na u al sys ems (Camb idge Uni e si y P ess, 2006).
23. Laaksonen, T., Hakka ainen, H. & Ko pimäki, E. Li e ime ep oduc ion o a o es –dwelling owl inc eases wi h age and a ea o
o es s. P oc. R. Soc. Lond. B Biol. Sci. 271, S461–S464, doi:10.1098/ sbl.2004.0221 (2004).
24. Mazgajski, T., D. & Rej , L. The e ec o o es pa ch size on he b eeding biology o he g ea spo ed woodpecke Dend ocopos
majo . Ann. Zool. Fenn. 43, 211–220 (2006).
25. Le To o ec, E., Helle, S., Suo sa, P. & Hakka ainen, H. Pa en al and gene ic ac o s a he han nes ing o es pa ch size a ec
physiological ai s o o sp ing in an a ea-sensi i e passe ine: an expe imen al s udy. Ibis 154, 345–353, doi:10.1111/ibi.2012.154.
issue-2 (2012).
26. Skagen, S. K. & Yackel Adams, A. A. Wea he e ec s on a ian b eeding pe o mance and implica ions o clima e change. Ecol. Appl.
22, 1131–1145, doi:10.1890/11-0291.1 (2012).
27. Hox, J. J. Mul ile el analysis: echniques and applica ions (Rou ledge, 2010).
28. An onakis, J., Bendahan, S., Jacqua , P. & Lali e, R. On making causal claims: A e iew and ecommenda ions. Leade sh. Q. 21,
1086–1120, doi:10.1016/j.leaqua.2010.10.010 (2010).
29. Finnish Fo es Resea ch Ins i u e. Finnish S a is ical Yea book o Fo es y (Finnish Fo es Resea ch Ins i u e, 2014).
30. Hanski, I. Single-species me apopula ion dynamics: concep s, models and obse a ions. Biol. J. Linn. Soc. 42, 17–38,
doi:10.1111/j.1095-8312.1991. b00549.x (1991).
31. Swi , T. L. & Hannon, S. J. C i ical h esholds associa ed wi h habi a loss: a e iew o he concep s, e idence, and applica ions. Biol.
Re . 85, 35–53, doi:10.1111/j.1469-185X.2009.00093.x (2010).
32. Suhonen, J. & Kui unen, M. In e sexual Fo aging Niche Di e en ia ion wi hin he B eeding Pai in he Common T eec eepe
Ce hia amilia is. O nis Scand. 22, 313–318(1991).
33. C amp, S. & Pe ins, C. M. Handbook o he Bi ds o Eu ope he Middle Eas and No h A ica Vol. 7: Flyca che s o Sh ikes (Ox o d
Uni e si y P ess, 1993).
34. McGa igal, K., Cushman, S. A., Neel, M. C. & Ene, E. FRAGSTATS 3: Spa ial Pa e n Analysis P og am o Ca ego ical Maps.
Compu e so wa e p og am p oduced by he au ho s a he Uni e si y o Massachuse s, Amhe s . h p://www.umass.edu/landeco/
esea ch/ ags a s/ ags a s.h ml (2002) (Da e o access: 13/02/2017).
35. Bollen, K. A. & Cu an, P. J. La en cu e models: a s uc u al equa ion pe spec i e (Wiley-In e science, 2006).
36. Hancock, G. R., Ha ing, J. R. & Law ence, F. R. S uc u al Equa ion Modeling: A Second Cou se (eds. Hancock, G. R. & Muelle , R.
O.) 309–342 (In o ma ion Age Publishing, 2013).
37. Yang, C.-C. E alua ing la en class analysis models in quali a i e pheno ype iden i ica ion. Compu . S a . Da a Anal. 50, 1090–1104,
doi:10.1016/j.csda.2004.11.004 (2006).
38. To ighi, D. & Ende s, C. K. Ad ances in la en a iable mix u e models (eds. Hancock, G. R. & Samuelsen, K. M.) 317–341
(In o ma ion Age Publishing, 2007).
39. Hilbe, J. M. Nega i e binomial eg ession (Camb idge Uni e si y P ess, 2011).
40. G imm, K. & Widaman, K. Residual S uc u es in La en G ow h Cu e Modeling. S uc . Equ. Model. Mul idiscip. J. 17, 424–442,
doi:10.1080/10705511.2010.489006 (2010).
41. Ende s, C. & Bandalos, D. The Rela i e Pe o mance o Full In o ma ion Maximum Likelihood Es ima ion o Missing Da a in
S uc u al Equa ion Models. S uc . Equ. Model. Mul idiscip. J. 8, 430–457, doi:10.1207/S15328007SEM0803_5 (2001).
42. Mu hén, L. K. & Mu hén, B. O. Mplus Use ’s Guide (Mu hén & Mu hén, 2015).
43. Klein, A. & Moosb ugge , H. Maximum likelihood es ima ion o la en in e ac ion e ec s wi h he LMS me hod. Psychome ika 65,
457–474, doi:10.1007/BF02296338 (2000).
Acknowledgemen s
This a icle is dedica ed o he memo y o D Ha i Hakka ainen. We hank H. Helle and A. Jän i o hei help in
ga he ing he ield da a used he e and K. G imm o s a is ical ad ice. We also hank nume ous e e ees o hei
help in imp o ing his manusc ip . This s udy was unded by he Tu ku Uni e si y Founda ion (E.L.T.), Academy
o Finland (g an no. 127875) (H.H.) and The Tu ku Collegium o Science and Medicine (S.H.).