Embed code
Note: this content has been automatically generated.
for inviting me uh mm thanks a lot for coming so so it's the first time that
i get to talk to a audience of computer scientists so i'm i'm sorry if
what i'm doing with maybe we it for you and that maybe is kind of
uh uh not the way you would two things about that in just a
ask questions or or interrupt or or or uh if if something is kind of um uh it so
i just give a talk us the way we would do that economics but but of course the
um i'm not quite sure how the audience will react right so if
if anything is kind of weird or unclear just just a interrupt
so that's what i'm gonna talk about is called um yeah it's a paper
is called a can power sharing foster peace uh evidence from northern ireland
and this is joint work with uh hundreds miller from barcelona and so the the question we wanna ask here is
and can power sharing uh be something that promotes a
promotes peace it's it's a tricky topic because um
well uh it's not easy to measure you know it's not easy to measure power sharing and it's not easy
uh to to establish a causal relationship statistically speaking
between these uh and uh and and peace
so um our first the start start with some examples
so i mean what what's the basic intuition here why do we expect that power sharing could play a role for peace
well the idea is the following imagine that that you are i'm a i'm a minority group in
society and you have a uh so to persons uh of the population or thirty five percent
french speaking switzerland for example french speaking switzerland if french speaking switzerland
was not re present it in the uh uh federal
council inbound if a friend speaking switch that was not we
presented as it heads uh of of department in administration
et cetera in the police force uh in all important uh a decision uh uh committees
what would happen well what may happen is that sooner or later
french speaking switzerland which would be a would be annoyed
and and and we started uh maybe uh some separatist movement right so what why hasn't it happened so far what why is there
no a french speaking a party the formal who wants to split from the rest of swiss and why why hasn't it happened
well the reason why it hasn't happened is that a friend speaking switzerland
uh is quite well we presented in the government at least always two to three people in the swiss
government are from the uh only from different speaking part of switzerland and at the same time
i'm also in in all the the important decision at all the high
level civil servants if if the if the boss officer is
is german speaking the the rice balls is a french speaking or vice versa
so usually you would always try to kind of have a every presentation between
lacked in switzerland and uh i kind of joe german speaking switzer
and so this kind of power sharing into its than what surely something that that would play a
role because they basically a a more general level what you would expect is ready for that
if a group is an important chair of the population is sort of the population but has no say in politics
uh well this group may have incentives to just to try to split from the country uh and
and this may end up in a in an s. so uh so that's for example um
if you think of rwanda well in one uh for example what you have is is
similar kind of constellation where basically the who two are awfully maybe seventy eighty percent
of the population you have to tutsi being much smaller and at the point window
who to took a uh took power at the beginning of the nineteen nineties
uh use use you saw that uh to choose what kind of completely marginalised
um and then that there was some to to to rebellion and and then uh you which a genocide
of uh of who to send a killing moderate to twos and uh and that
switches with one one million people dying in in in within hundred days
so so clearly the this whole thing is quite a explodes if it's quite that
important question of how you accommodate different groups and how they can share power
um and so so well it's it's okay idea is really um as malcolm x. was saying it's
the the ballot or the bullet either we can cost of this was about a black americans
either we we can have real representation we can cast the pilot or if we cannot cast about well the
only thing that remains is the bullet and so so that's kind of the the trade off here um
uh so it's really a kind of comparing the relative gains of sticking to electoral
politics or leaving electoral politics and going for for for some armed rebellion
i give you two examples of where power sharing a one can argue made a
really difference first thing is to switch civil war and will view is swiss
and while viewing knows about the this with civil war
okay so a civil war was in eighteen a forty seven eighteen for forty seven it
was between the catholic uh cantons of switzerland who wanted just to lose defensive alliance
and the protestant cantons of switzerland wanted to build a nation state as it was kind of
uh invoke a in this century and so basically um the protestant cantons
one um and so they have the choice a option one was
the uh uh kill the catholics across ten option to was to do some power sharing and and
they went for option two and option to what they did imply what it implied was that
i'm a very rapidly the catholic conservatives uh at a ants it's it's fall
apart partied a a p. d. c. uh was kind of strong invalid
and so so the cuddly conservative party would kind of christian democrat party
would kind of get very rapidly access to the to the government
at the same time uh a parliament which to chandler's a re
presenting giving a low at blocking minority to this small
kayak aptly cantons would tend to be small so in switzerland you know oh around lucerne most of the
small cantons or actually catholic which means that in terms of
blocking a majority of the double majority of uh oh
oh floats in the parliament but also for the direct
democratic elections actually the small cat leash cantons uh
close to loose and they actually have probably even a higher weight then the
protestant cantons and you even see that today in some popular vote
so what happened is kind of revolutionary the protestant winners of the battle
instead of crashing the opposition they actually almost gave more
rights to the opposition then they kept for themself
so it was power sharing which was not just for but going maybe even be unfair
and so that's that's kind of great what happened as an outcome was that well since
then we one of the richest countries in the world and a piece ever since
now let's take a look at northern island norton island what happened there on on an island historically the
protestants were kind of a dominating the state and the catholics in sixty nine they started kind of
the civic rights movement to to kind of uh uh uh fighting back
no i'll talk in more detail about the whole uh the history of it bit bigger perspective
but basically a pen for thirty years the protestants didn't want to share much power
uh and there was this uh um a paramilitary fighting from the catholic side and in
ninety eights they decided to make a power a sharing deal
so i brokered by twenty player um that it's
called the the belfast agreement on good friday agreement because it when it was in is the time
and so in this agreement what they they shared power in the government uh with with catholics and protestants in it
in the administration in the police force everywhere in society uh the catholics got their share
of power and since then um uh acts of violence have rapidly plunged okay
um now this is only one data point right we are the two anecdotes and and what we gonna do in
this paper is much broader than that so we gonna look at that micro data for for little districts
so we're not just it's not the paper with two anecdotes on on on two events it's like with many more data points
that that the problem of course is it is it is a methodological problem here because if you know yeah if if they
would just say well if you have power sharing uh and
after power sharing um you you have a more pace
uh is that correlation or causality well it may just be
caught correlation is spurious correlation because imagine if if i
if i share if i share a uh um uh i see a car which are well you should we
share the train but if we come here with john i share a car and afterwards we we have a
bonding and we get along well together then is it really that the fact that we shared a car
made that we get the long well together or is it the fact that
we got where uh along well together which was in the first place
making us share a car right so if we would just say the fact
that you have power sharing and then you get along well together
the uh the the doesn't prove any causality may just be pure correlation
it may be rather be getting along well together that would
determine the power sharing and what we're gonna do with this paper is we try to tease out this correlation you want to show that
actually some random changes in power sharing can show
a call so impact of power sharing on
uh getting along well together right that's that's that's the the menu of the day um
and so basically what do we do when this paper what we do in this paper
is an empirical analysis of the impact of power sharing at the local level
so we knew was data from northern islands twenty six local
district councils so it's like twenty six cantons uh
and so the observation is a district council and they here so it would be
like if we take the twenty six swiss cantons for the last thirty years
and so uh we we we we get a at a thousand data points a bit less
uh and so for each uh for each uh of these these
three years we know whether there was power sharing or not
and now the the slide okay um and so what we do is is called a and three a first we do
um fixed effects regression ready filter out all the the the normal to the time invariant
uh saying is yeah in these districts and we just look at what happens after the after shocks if
power che uh if power sharing changes does this trigger it change later on in in in violence
and then we do something else which is called instrumental valuable regression very basically
exploring to random variation in whether power sharing a takes place or not
now i don't uh uh in terms of the audience i know that
informatics since globally on everything that's technical their way by that
and uh in all the disciplines but maybe the way you do things maybe different that that then the way economist two things
so who will few works which data sets like that who would kind of two
statistical analysis we crash analysis of they visit who of you would do that
everybody so everybody of you would to uh but but then they're also informative since we would uh would
uh right right algorithms and to them by but you do both actually everybody does pose kind of
okay very good and so everybody of you knows whether fixed effects is or not or
you make the call that differently or who knows what to fix the fact is
so what okay what what we call by fixed effects is really just a fancy name for
for something really simple is the in economics we like to do that we take something
a really simple and then we give it a fancy name and we feel smart so so so
so basically it's like a separate constant for for each each group so if you have um
um is like can i write on this whiteboard here or not ah okay so if i i with
this i would actually destroyed if i would write on it right and you would get into trouble
so so basically yeah so so we all know so this is to find this or
okay so you know we all know what what what the real regression is
so we have this uh points we have here um invaluable uh aches
that could be power sharing p. s. and what we wanna explain is uh a piece
or maybe you know the peace symbol right uh and so we want to explain that and so
we have this data points and we we try to find a regression line that would minimise
the the deviations from there it's the best fit kind kind
of the uh minimise the least squared deviation from
from this uh from this line now uh upper what fixed effects would do is would
for four separate categories of points it would uh it would kind of um
allow uh here's kind of a heterogeneous intercepts so the intercept here
would we would allow h. to this constantly would allow it to be different for different categories
so if we would run repression um of maybe a how does cooling to or what's the
impact of schooling on i. q. uh or what no sorry what's the impact off
of schooling or on on performance later on and we would do that and and women are kind of
smarter right women are are are smarter and so their performance to start with would be better
so what we what we would then have is that we would we would look
at the impact of education on performance but we would allow for different intercept
if the if the women are always ten points by that and the man anyway
so that the constant uh for the men maybe ten and
the constant uh a constant for women maybe twenty
and so so by allowing it for different constant for different subgroups
what we do is we filter um on observed heterogeneity if something is different about
women being smarter then man and we cannot really captured without control variables
then this different constant would kind of filter out for that right
um and he it's exactly the same here what we tool uh integration this that some of these districts are different
some of these districts have a history of getting along well together and and some of these
this three streets have have a have a a tough history where they have been already bottles
in the middle middle ages et cetera and so by by by allowing for different constant
uh what we do is we filter out all this time invariant invariance uh differences and we just
uh uh compare at the impact of these shocks that that's what we do statistically
and then and then we do other stuff statistically but all gonna um
yeah i'll gonna talk about that uh when you get there so that's what we called fixed effects basically
and then the the closely uh elections with the idea explain when you get there
um so what do we find well they find is that power sharing does indeed curve fatalities
and here we find that a call so the impact so it's not just correlation it's not just
the guys getting along better would kind of to power sharing it's really that once for
a for random reasons in someplace power sharing starts in in some other place power sharing doesn't
start well this would actually trigger a decline in violence um and this effect would persist
as as long as power sharing is in place but it would fade away once power sharing is going
what does that mean for switzerland what it means for switzerland if if now some crazy party would get
uh into power and uh and it get rid of the magic formula and kind of uh would
kick out all french speakers from the swiss government well we may well in five years have
some says a secessionist movement big trouble like so it means that power sharing in order for
power sharing to to war and peace well it needs to remain in place you cannot
just have it for a few years and then you remove within ten the effects would be perfect
assistance that's not the case we needs to stay otherwise the fit the good effects are gone
okay it's yeah uh and that the effects of
quantitatively sizable so northern island um being um
not an i love being really uh uh i was were very tough conflict with several thousand people dying
we're roughly two thousand people dying and and here we can
explain basically uh when you have power sharing the
uh without power sharing the effects um the the risk of of a fatality
would be roughly three times bigger and so if you would uh
completely uh if you had instead of having power sharing in some places if you had
had full power sharing everywhere that would save a town that lives in this context
so quantitatively uh if we would go for full power sharing for for the whole period in all
uh in all um in all districts we could kind of cox uh fatalities
by half so it it's a huge effect um now maybe uh
something other
maybe before i go to related to what you may find we're to which maybe in natural science but people
may find where is that what what what the heck does does this guy focus on north northern ireland
why don't you focus on the whole world or something like that well the the the the the
thing what's very special about northern ireland what's special about not an island is not an island
is is a rich country uh with conflict and that's very rare usually rich countries don't have
conflict uh uh i mean or or or what you find in the literature that is
uh is wells uh uh prosperity is one of the biggest predictors of peace
statistically speaking so you have a few rich countries with conflict think of past country think of northern island et
cetera but they're not many and so it's very rare to have a country that has extremely good statistics
a statistical capacity very detailed information and at the same time
uh really a plot the conflict that that that
that's important to explain right uh and so uh that's why this makes uh not and i'm quite unique
and the state um actually this this uh to take that we get from northern ireland
we wouldn't be able to get for any other country in this in this for
um and if if we had less uh if we would do that on a on a more global scale the problem would
be that that uh we you you couldn't really move you
want a correlations and economics uh what we kind of
we we lie uh so it's always in a committed solicit a trade off we perceive between
being as tight as possible in terms of causality in explaining causality not just correlation verse who's
uh external validity and a covering the biggest part of data and of the world as possible
here we we go for the for kind of um one country
but with very good pay a with relatively good data
a while in order start this uh if if you're interested i can talk also about my all the research
we would sometimes cover the whole of africa for example so we also papers very quotable of africa
here i'll um so i have until until a lunchtime right
that's well or until noon so so basically here i i may also give you
a a bit of an overview of what other people do an economy so
some maybe maybe you don't have so many economists coming over
to give talks right so so basically economics um
i i make religious conflict is kind of a and fixing a white
well actually we're uh so if if i talk to my
first year students about conflict they always say but okay why do
you talk about political science and economics is that economics
well it is economics because globally what what people have found is that in the fifty poorest
countries of the world in terms economically speaking well forty out of the the fifty
at uh are the main problem is political stability so if you want to explain development
a few well if you want to do development economics explain grows and prosperity
well the first thing you should look at is not sings like inflation and
the classical uh economic factors are you what you should look at
one of one of the foremost factors that would create poverty is really political instability
political instability would be a huge uh you tracked and that's why uh in the in the last years the the
literature has really been exploding it's a huge boom of studying
and things like political instability in conflict in economics
and so it make uh ethnic conflict is like a big big topic so there are many people
starting it um what people would focus on uh one of the the results which is
this is not kind of a hundred percent related to my paper but i just
save for the for the sake of of of of uh it's important saying and and and given 'em but i give a little
a only of the leadership one concept that's really important is eight
make polarisation and so these people here for example um
as the one my rather right right uh or they're not get all the methodical with the
people who have studied this and i think polarisation is you craig in this is off
of composition of ethnic groups in in the in the country and what these papers have found
is that actually in a country that's completely ethnically homogeneous think of norway um
you don't have a huge level of comfort usually in a country that has
a home that's off small groups think of car now you don't have many
conflicts neither what is really the dangers composition what they found is
two uh two or three really big groups of similar size and that's the
the polarisation index and polarisation index would be maximum would be uh it's
in index from zero to one and it would be close to one if you have two groups of almost fifty percent of the population
and countries that would to uh would have very high polarisation would would be a
condors like angle are quite them allow one double room the uh all
these countries with with uh where where you see in the data have had
a lot of conflict a country where polarisation would be very high
but has been peacefully switzerland and that's why our institutions are really important and that's why i kind of
a a in this research focus on or whether is it possible to export nice with institutions
could that also working another country is kind of swiss uh a tradition of power sharing um yeah
institutional conflict again there there are a a series of uh of of papers on that
usually that what what what the papers find is not likely a very clear cut effect of um
yeah not a very clear cut effect off democracy uh what
what you see is that a full democracies like switzerland
uh they usually tend to have a no conflict or very little uh
then full autocracy is like north korea and they would uh uh
this film tried a full full apocryphal autocracy is like some conducted a really
full autocracy is well in this full autocracy is uh an again
i in full autocracy i it's kind of a level of conflict is
not very high because uh it's really hard to organise any
any uh any uh a manifestation demonstration without being put to jail
or while kind of that at the places where where i'm a
complete is highest people from the r. and all classes
and across seas which uh which uh is a concept saying countries in the middle
where you have some democracy but not full democracy think
of a venezuela for example venezuela you would have
a uh we would have enough democracy that people could go in the streets and organise themselves
but it's very far from a full democracy which means that
people have still grievances and complaints uh to formulate
while uh they are still have the means to convert to complain while in the full autocracy
spate they have a lot of things to complain but they they don't have the means
to complain while in switzerland they have the means to complain but not much to complain about
right so so it's kind of in the in the middle range where its dangers
another result in in a paper have with paul paul collier uh
is like uh what we find is that democracy actually um
the market is usually would would reduce a a complete a lot in in the really rich countries
while in in poor countries uh the uh it's it's a bit more ambiguous because in um
i had it that there is this dangerous phase at the moment when elections take
place so there's quite a lot of electoral violent think of cain yeah
uh and so so basically uh one of the conclusions is that
especially if if if in the country uh to to to take the risk of conflict on the
line would be relatively big and there's a lot of or the rounded all these risk factors
their democracy is good but you should try to accompany it for example with peacekeepers and observer
independent observers et cetera so it's not like you we cannot have just the naive
a ration of maybe george w. bush uh some years back or saying
well let's organise elections until gonna have peace or maybe seen uh in afghanistan and iraq it
it's not just the fact of running an election that would kind of automatically deliver peace right
so this would be or the or the results here in this literature now when we talk about um power sharing
they talk about power sharing our like more specifically what exists uh as paper some power sharing
i'm kind of the existing guru on power sharing in political science would
be lost eric sediments so we would have a few papers
where he would uh have expert coding of pay they would go for it to a country would sell we
sink in this country um peace groups are kind of sherry power and how does this compare with conflict
now this is this is great because this is already much better than what we had before
but it has a few problems to put the first problem is it's
hand coded by experts so the experts of course if they're coding
um well they are influenced by how well uh groups get get along well together
so if if you know if groups are unified stand that would maybe
prevents them to coded as power sharing so the coding is androgynous
that would be the coding maybe uh there's cognitive bias so
you would code something uh influenced by if you
caught something in your influence by the results then you would have in her independence of the oh
if they get along well together no if there's peace well we call that power sharing if it's not
peace we don't call the power sure so it's kind of the decoding of the variables problematic
that the second thing is also again that there is there a in
this state or not uh some random shocks in power sharing
so it's the near correlation and the mute correlation again as i
was saying if uh john and i take the train together
uh and then uh have a good laugh and and worker on a paper
with phil uh only the uh uh if uh we have a good
laugh then it's not the the train that has great it's the that
the collaboration but it's the collaboration that is created to train right
so so so kind of um yeah so that that's a bit and and the
data is really a aggregate so what what what do we do differently
um uh here is well uh we can only do it for one country because for the rest of it wouldn't be there but
but for one country we have they that's much more precise than what exists in all the in the existing with the chip
we have also data on the perpetrators and the big been so we really it's not just like i carry gets
people dying and we don't know why they die but we're actually no for each uh killing we know
who is the perpetrator uh from which religious background is the provider who is the big
them what is the reason the purple white upper uh the person was killed
and so it's really uh it's for for each uh event of the of
these two dozen events we have really detailed information about everything and um
especially important is that we code power she'll uh sharing from administrative records
so in the u. k. what what they have is a a a freedom uh of uh information though
uh the freedom of information little is is something great for for researchers because
if you send an email to uh to l. a. u. k. government a unit
i am um and if i mean if you if you don't ask them to
work for free for months for you then if you if if if your
if you request can be dealt with within half a day or
so then they are obliged to on so you for transparency
so if we we have sent to emails to all of these twenty six a district
saying we want a list of all the the mayors to heads of these local
government and the vice mares the that the advise the president of these local
governments for the last forty years uh you have to send it to us
the core research more politely uh uh with the freedom of information little
and so uh most of them send it immediately and then some we had to send like
uh ten emails and have asked them a bit but at the end everybody answered
and so uh the law requires them to answer that and so we have of of these
of these uh of these districts we have a bit of you know who are
awarded the two main person in the government and historically how it works in northern island
the power sharing is that if you have power sharing between to go to parties
uh what you typically would do is some rotation involving that the head of the government and the vice set
so if we have a power sharing between this way since the
and the south africans uh for example then it would be
that uh once they had a south african and once the the vice say this within the next year we turn around
and so if you want to identify a some explicit or implicit power sharing
in in a government there what you can do is to see if
the if the head is catholic uh of catholic group uh and
and the vice head of a protestant party or vice versa then that would be
typically uh and power sharing it's of course may should with error but um
but it's automatic coding with another retirement automatic coding it eliminates all these cognitive
bias that you would have if experts would coded so we kind of
like this uh automatic way of coding uh and the
error to air it gets the noise where we
wrongly attribute something to power sharing or we we miss out on power sharing well we would
think of that most there's at integration by us so uh at
tenacious bias meaning that if in these regulations you add noise
well it makes the it makes the results we couldn't they are actually are
in reality so it may on their estimate if anything is the length of our results and so
and then we do this fixed effects which again but people don't to in the
literature so far uh on power sharing and then the last point is
that we yeah we have a we exploit exogenous
variation around the majority special so want me
to actually then in the second part of the paper is we compare uh cases
where um uh that that the biggest party has
is close to the fifty percent threshold but doesn't reach it with cases where
there really similar but that that that part is just a ball
the the fifty percent thresholding can govern alone so we for example uh if for example you
you are a party that that that reaches a forty nine percent of the votes
um and then you oblige or more you either you do i hung a hung government
like with a minority government or you have to reach out to the other group
to to form a coalition government like mercury in in germany if it works out a you try to do
a a um a coalition government uh and um so if you have forty nine percent of the votes
you are really kind of uh you have big incentives to do a coalition government with the others
power sharing while if now you have fifty one percent of the votes are very compatible case we have
fifty one percent of the vote well then you can govern alone you don't need the others
and so we're we we don't want to uh compare apples with peers but we want
to compare things that are comparable and so if we compare it the the
the the co consolation which forty nine percent of the votes with fifty one percent of the vote
it's just random whether you're at forty nine or fifty one but with forty nine you may
not to the power sharing well with fifty one you may do the power sharing
and so that's a random variation and make such as variations we would call in economics
that would kind of trigger to power sharing in so we can identify the causal impact
of the power sharing own then the the outcome of of peace or no peace
so that's kind of to distract the chain that's also update allows to do that uh up to some point
while the existing data out a pair of you people before wouldn't allow that right
so that's a no another kind of selling point of what we're doing
uh a few words on the on the context of northern island so norton islands so okay i don't know why uh um
yeah yeah project i didn't like the context of northern ireland so that so uh
northern part of island poster has been religiously divided since the sixteenth century
in nineteen and and so that has been kind of trouble uh of course uh
between the two different religious group in order and i in ireland in general
for a long time then i learnt became a country to probably go file in the beginning of the twenty century
uh but then things calmed down a bit and a violent again broke
out in nineteen sixty eight nineteen sixty eight was with the global
civic rights movement all around the world you know like uh the older hippies and stuff
and so basically i'm from sixty nine onwards sectarian violence exploded
um and and you had first or peaceful demonstrations and then if you would know this song of you to
a sunday bloody sunday uh so that was like um a demonstration but and the the the soldiers
we kind of shoot in in the in the in the crowd and that would
be a big list of the violence because it was perceived as kind of
uh arbitrary nasty state violence right uh yeah and so so kind of uh
um uh yeah and and the perception i mean
historically speaking i it was like a terrible
i tell people reaction of the of the police forces and so basically uh from
then on a while inspired so violence would explode from sixty nine on wood
and um and uh uh at some point uh even parts of terry or belfast
the police wouldn't even they're entering they were kind of no go zones
and then from seventy six it would kind of become a bit more quiet also because
the u. k. would send more troops in so it was more like that
they're saying okay by sending more troops in they could kind of get better control of the situation but it has to buy no way
sending troops in alone would not solve the problem and the problem was really kind of souls
in ninety eight where uh you had like on the national level a power sharing
on the national level between catholics and protestants but all ready before that
in the nineteen from the nineteen seventies on which they were local power sharing
at the local district eleven so it was kind of the it is experimentation on the
local level with power sharing that would then prepared a fertile ground for a nationwide
power sharing in ninety eight and what we can exploit here in the paper is
stays in the local power sharing in the in the seventies eighties and nineties
and not like the the topeka nation a national event because the big
national event this one data point we cannot do regression with one
data point right so so yeah so that's that's that's what we
what we gonna uh focus on now of course if um
now would break six troubles uh are again flaming up a bit so so
if you read for example recent article in the economist of last week
uh again um is so does that their travels to form
a a government and parties don't get along uh uh
uh uh well together at the moment uh and so the big problem is of course that that uh
and the question of what happens with with norton island uh uh what was not
really important as long as uh i was less import less salient as
long as the u. k. and ireland repose part of the u. because the
the border between rip public of uh i learned the northern island
what was a very porous border because it was posing the you know if the u.
k. nice to you then of course it becomes a hard border again between north
and south island and then that this may trigger new you trouble so so uh it
it's clear that uh that that it it's not over although since ninety eight
things are much better so our our data and fatalities data from stockton and kane
and we have all it in another paper with with thomas miller into features also
we have a created a and a data sets that are based on this available data very it's very
chill localised so you know the uh latitude longitude we
know exactly which treat as things take place
um the power sharing is collected by us as i was saying that the way we call what we call this power sharing is if
it off to to uh uh main people in the government one is
of the catholic block and one is of the protestant block
and we also to various refinements so we we know that for example some parties they were hostile to power sharing
so we would also say okay now let's filter let's only caught a
code uh this is partially if you have the catholic block
and the protestant block and none of the parties involved are one
of the uh of the with the ones that are notoriously
house style to power sharing and so we we do like three or four different ways
of coding it and show robustness to the to the of the results with that
and then we have various results on from that sense was on the share of catholics protestants et cetera
so in terms of what what what is the role correlation before doing any regression let's have
a look at the road data here you see the the the correlation or basically
oh here we have the timeline horizontally vertically you have
the number of council districts with power sharing and
that's uh that the dotted variable and so it is a rare is from zero to twenty six
uh so if everybody would and so it goes up to eighteen side at the
at the end quite many would actually do the power sharing while um
uh here the total fatalities that's the the solid line and that which vary from zero to forty
uh you see uh so that would be good people being killed in it is strict in
here so that would be a would be a fatalities in the cool poll a
the rule uh in the here uh uh seventy five for example so it would be one district a a
or cantonese with everybody would be a and here so it's it's quite scores of scented
numbers what you see is the role correlation is kind of negative meaning that
when you have low power sharing of more violent and where you have high power sharing you have less violent
now again this could just be a correlation so that's why we have to go further but that's like the big picture
and so now let's let's look at what happens with uh the regressions so very questions um
so yeah uh you'll you'll know what what s. is right if i understood well right right okay
so yeah so we start with with with this it's simplest possible regression that well as
very basically put a does this fit uh uh between
the the the the the datsun b. we minimise
the squared deviation from from the fit line so we just try try to find the best fit
that that the observation is the district here as i was saying i can't and the uh
um the dependent variable is the number of complete related and vitality spurt thousand inhabitants
so uh so that that uh yeah it's it's
per population in a given administrative district here
the main experiment about it was a dummy of whether there is power sharing or not
so uh whether the little power is shared or not and so that's the regression read on so the equation is the following
so we explain fatalities at the district here level it's are
dependent variable that explain explain it with with a constant
and so actually with the fixed effects we gonna have constants that are specific for each cantons where um
then we we explain it with uh with our valuable power
sharing which batteries at the district here level so
uh it can belfast that you know prefaced would be several districts but
but like the western belfast district in a here seventy six could be power sharing but in seventy seven not
so it it varies at the district here level then we have uh this fixed effects which is a constant
for each district and we also have a et time affects which is the constant for each year
which would filter out every a global shocks hitting the whole economy so if
for example a closer to the big power power uh a sharing
uh later on the the player uh agreement of um
the good friday agreement if if getting closer to
the years uh two thousand two thousand well if that is something that would kind of um
that would uh decrease violence altogether well then this global effect hitting the whole
lot an island would be a a fill that out by the fact
that we also have a a a dummy variable account at a constant
or for each year which which with again a filter these out
and then we have some some uh some vector of all the control variables bearing
at that year level uh like a district your level and some editor
um so what that these are the results for for this or let's regression so uh the left hand what explains
casualties per capita how many people die per hundred thousand uh uh at the variable of interest is power sharing
first we we just do a analysts without doing these fixed effects in this
time effect and then with fixed in time effects uh um that
specification and then we add some controls we control for this teachers of catholic parties of
the protestant parties and and that's the the result what we get is basically
well let me get is uh that we find this statistically significant effect that the
one percent level of power sharing reducing that the level of conflict and
and here uh uh well that the corporation doesn't say so much but as i
was saying effects is really big if if you uh if you um
i it would actually given at the part baseline risk is low uh
uh here uh it would uh it can full power sharing would there would uh
the two three times less uh fatalities then if you have zero power sharing
so that's kind of what we what we do first and then we we look at okay and so uh intuitively
when we run this regressions we can estimate um we can
estimate what would be the average level of cat casualties
uh twelve me in the twelve months before power sharing start
and in the twelve months after power sharing start
so that's just a visual that's the same saying more or less than the regression that
i've just shown us the table but just a visual representation of that issue like
and so what you see here that that's at the level of says you on what you see
is that before the power sharing starts fatalities are are larger and after our saying starts
about these are lower right yeah that's so that's what you see visually and then if we
look at the what happens if you stop our sharing it's the other way around
so meaning that if you if you stop power sharing here at the point zero and you compare the
twelve months before with twelve months off there's and here on the scale of the casualties per head
what you see is that uh what while power sharing is still
in place casualties allow and then off the power sharing stops
casualties go off and that's what i was saying is that well
if now in switzerland we we uh we we we had
we had the bad idea of removing a whole lacked in speaking
switzerland from the government uh very soon troubles could start
so it's not like something that you can take for granted you always have to keep up the the
power sharing otherwise it's beneficial effects uh disappear and
that's what you see here graphically now um
let's uh let's talk so uh i uh let's talk a bit about uh the
this ad to s. l. s. a regression so well viewer knows what to s. l. s. it's
so it's the same thing as instrumental variables kind of it's a two
stage least square aggression meant maybe you do exactly the same
but you call it differently the kick it may well be but basically what we do there is we have a first h.
and in the first stage we we explain when power sharing takes place
and then um uh i can also do a show that graphically of it
uh so basically what we have is we have um a viable
majority so we have this variable this valuable no majority uh uh
would affect
the variable
what what uh affect the valuable power sharing
and the valuable power sharing
worked effect
a conflict the numbers of people killed right
okay maybe yeah i know that
but no majority shouldn't have a direct effect
on on conflict and and so it's it it's uh should have a direct effect on conflict
um when you control for vouchers of the of the of of the parking that's what what what we do
and so so basically what what does that mean well what what what we do is we
we want to see what's that we want to see what's the the the in what's the impact of power sharing
on conflict but the problem is that if we don't use this instrument of no majority well that could be
a a causality running the other way around that conflict eaters power
sharing and so given that we're interested in the cool also
effect of power sharing on conflict but uh uh and want to
filter out any all their effect going in the other
direction what we do is we we it was a variable
that affects power sharing but that doesn't affect conflict
and so that that uh uh so so that's uh that that's the idea behind the
idea is that we want to look at at some kind of a random variation
uh in in the power sharing a valuable um that would not have a direct effect on conflict
uh and here again as a saying if you're close to this threshold if we
compare that in a forty nine votes per cents with the fifty one percent
we don't we of course if we compare a place with zero percent of
protestant boats with a place with a hundred percent protestant about these places are different that's
clear so we cannot compare them but if we compare forty nine which fifty one
vote vote chip then the places are very comparable and we expect
uh the the no majority valuable tool only affect power sharing
uh uh as complex rule the channel of power sharing and that's kind of the
idea of what we have here um and so when we do that yeah
when we do that maybe to show first the first stage that what we do in the first h. at the first stages this
and then the second stage uh is is the strike that's the second
stage and that's why collect two stages critically deaths to state
so the first stage we look is no majority really a powerful uh an explanatory variable
of power sherry does it really matter in what we see here yes it does
so if not always sectarian party neither the catholic or the protestant block
has it been that they're also neutral parties that are non sectarian
but but if neither of the of the two blocks has a majority well this would what's the likelihood
of power sharing by almost forty percent so there is also power sharing that takes place if
if it's not a if a if a if a a group as a majority so even if you
have to fifty one per cents you may still be nice enough to do the power sharing but
the likelihood in terms of pro probability that power sharing takes place is
much larger if nobody has the the the the majority right
and so that's that's the first thing so so we find indeed that no majority
the powerful instrument of power sharing an l. let's look at the second stage
if we look at the second stage what we see is then that's that this
instrument of power sharing a that is a in the c. i. decreases conflict
very substantially and uh that's what we see here even if you control for the teachers et cetera
uh for the fixed effect so we continued to control and then me what we do is we look at different band with
so here we compare it situation uh where turbo chairs of
the biggest part is between thirty percent and seventy percent
a bit close to the to the re we cannot get as close as
forty nine versus fifty one because you wouldn't have enough data points
but first we we are rather browse between thirty and seventy
and then between uh thirty five and a sixty five
and then at the closest we can get is really between forty
and sixty percent of of work so ten percentage points
from the majority threshold where we argue that indeed the
yeah a place where you get forty five percent or fifty five percent of the vote are still kind of comparable
now of course uh if we wanted to do even better if we wanted really to kind of
um uh i uh i i if you wanted to be
even more confident on the results what we would have
to have would be more data points so that we could have a a even more narrow window right
um so so by the way something i forgot to stated that the paper be prepared it
for a for a for a journal called economic policy in case uh
uh so we eh revising its uh i currently for for for for this
for this journal which is just something i i should also say uh for the
top uh yeah so so basically um yeah so that that that's the perfect
we find that that indeed even for a narrow bandwidth even when you control for all this stuff
we we find this this causal effect of power sharing own conflict
um then we do a a bunch of robustness checked well not gonna talk about them in detail
so we use different population data different ways of coding power sharing it different construction of the boundaries
uh we also stock in ninety five uh in some specifications
just to be sure that that we don't pick up a nationwide uh effect of that big ninety eight
power sharing nationwide we also restrict yourself around election right before the election right
after the election because uh for the control variables lack causality is
uh or for example the demographic controls or political
uh uh controls so um yeah and then
we do heterogeneous effect but that's basically don't find anything right uh is uh nothing is significant so we we
wanted to look at to be defined a stronger effect if there are more catholics in a place
and we don't really we find a an effect that speaks in other no
matter whether you have more or less catholics in in a place
so to conclude um so that this or ethical argument white power
sharing could matter is is that it's very simple but
it makes sense right to say that if you're completely excluded from power well you may have incentives
as malcolm x. puts it to to uh to replace the the ballot with the bullet
i'm not an island is an ideal setting to study this because a kind of a it's
a rich country having conflict and they're not many cases of rich countries are in conflict
so the data quality is excellent here compared to what we could otherwise get
um the empirical strike what what does it do that is it is it links to me to
uh so uh empirical strategy other saying based on fixed effects
instrumental valuable and restricting good observations close to this
rational conclusion is that power sharing reduces violent
not just a few words on the cabinets on the on we have to be we have to results of course as always have
to be taken with a grain of salt so what what are the things you have to be careful about first of all
that weakness of the paper is that we don't have so many observations right next to this
racial so i would have loved to run the regression just forty nine versus fifty one
and so given that there is not enough data points we have to have a a a a
window that is that is not so small so we have to we have to buy
if we if we be in the even the results have to buy the assumptions that place
with forty five percent of words and fifty five per support are not very different right
that's the first copy at um while if we if we uh if we have more data we could do something
which is called economics a regression discontinuity design i don't know if you do that also come you designs
this is quite a cool the methodology but basically you would really compare
uh that's the threshold uh uh to the different um at the
different observations uh and you can also do that graphically this is quite cool but we don't have enough alterations to do that
second a caddy at is external validity is it of course we have one country
and what works for switzerland and works for northern ireland does it mean
that it also needs necessary to work for libya or live along
uh or three long car well it's not clear right the context to really really matters as well and so
what what would be really important especially if you think of uh uh for example e. d. i.
it may be harder to do a power make power sharing work that they've been what is
in switzerland it and in a northern island so what would be really interesting now
if if somebody could do some similar a paper but with data for for ever come
to that is very different maybe from from another continent maybe confidence less rich
so that we see that how how far to results can be generalised so that

Share this talk: 

Conference program

Recommended talks

Medium Voltage High Power Electronics
Drazen Dujic, Power Electronics Laboratory
16 March 2017 · 9:50 a.m.
History of Swissmetro
Marcel Jufer, Professor
1 March 2018 · 6:17 p.m.