00:05
welcome everyone uh
00:06
this is uh part two so uh on advanced
00:09
gis method training
00:10
so aha multi-dimensional indices and
00:13
and this in this part we're going to
00:15
focus on the the methodology
00:17
or the kind of set of methodologies of
00:19
multi-dimensional indices and how we use
00:22
that to create
00:23
the our heart index if you haven't seen
00:25
part one already
00:27
go and watch part one first because that
00:28
explains what our
00:30
bar is and how it's created uh so it's
00:33
it's important that you've seen that
00:35
already
00:36
um with me today i've got uh
00:40
dr mark green and dr costas daras
00:43
who are both involved in creating the
00:46
arhar index
00:47
and so i'm going to take us through a
00:49
few slides where we explain
00:51
how we created and mark and costas are
00:53
going to give us some more details about
00:55
how they did what they did and why and
00:58
some of the kind of pros and cons of the
01:00
decisions they chose to create the index
01:03
in that sense
01:08
so you know we've kind of got the top
01:11
question of
01:12
how was our heart created so we've
01:14
already said a little bit about
01:15
the different domains and the different
01:18
indicators within those domains
01:20
and some of the data sources that we've
01:22
used
01:24
and just to dig into a little
01:28
bit more of the the data set
01:30
specifically so
01:31
there was a selection of the retail data
01:34
was what we might call consumer data so
01:36
a lot of this was from the consumer data
01:39
research center
01:40
in terms of business locations business
01:42
types
01:43
and that set but the majority of the
01:46
data
01:47
was open data from a variety of
01:50
different sources
01:51
a lot of the health related data was
01:53
from nhs digital
01:55
nhs scotland for example hospital
01:58
locations gp locations
02:00
uh open street map uh we use quite
02:03
heavily in
02:04
the routing element the the routine
02:07
section which we'll
02:08
get to in a bit and we use uh data from
02:12
open uh ordnance survey open data green
02:15
spaces
02:16
um and the european settlement map for
02:19
population information
02:20
and uh postcodes and lsoa were the kind
02:24
of
02:24
two key geographies uh used in this data
02:27
set
02:28
um as we're working with spatial data
02:31
we'll end up with a lot of these terms
02:33
so
02:33
if you're not familiar with lsoas or
02:35
postcodes there's lots of resources out
02:37
there too
02:39
and to give you a bit more information
02:41
about those
02:43
so um i just want to ask
02:46
kind of mark to start with
02:49
why did you pick these data sets and
02:51
what was the kind of process
02:53
they're very broad level
02:56
so the reason we stuck mostly with open
02:59
data is that we wanted the entire
03:01
process to be
03:02
replicable so you could anybody could if
03:05
they wanted to recreate our resource
03:07
recreate our
03:09
products they could access all of that
03:11
data
03:12
and i think one of the kind of broader
03:15
discussion points is around kind of some
03:17
data needs to be held back for sensitive
03:19
reasons
03:21
um but we really want to be trying to
03:22
make that data openly accessible so that
03:24
anyone can get hold of it it might be
03:26
that
03:27
you know a local government's really
03:28
interested in looking at our data and
03:30
access to gpa
03:32
and services and then they want the more
03:35
specific
03:36
locations as well and so it's i think
03:38
it's important that we
03:39
where we can have open data that are
03:41
replicable
03:42
that people can get hold of and see our
03:45
process
03:46
and that we're not trying to hold hide
03:48
anything behind any smoke and mirrors
03:49
really
03:51
okay great and cost us um
03:55
in terms of selecting those specific
03:57
data sets how did you
03:59
come up with that list in particular
04:03
i think that was based on the
04:05
theoretical
04:07
basis of this index so our idea was to
04:11
create an accessibility to health
04:15
assets and hazards index so you start
04:18
with this point
04:19
and then from there you're looking what
04:22
data are available then you're looking
04:24
for the
04:25
quality of this data and
04:28
having decide which geography
04:32
is going to be the reference geography
04:35
for your index
04:36
and then you can make the decision if
04:39
this
04:40
data are appropriate or not so
04:44
yeah the reference geography being ls08s
04:47
sorry
04:47
the lower super old material so it's
04:49
kind of what
04:50
level do you want the index to be and
04:53
then
04:54
how what data is suitable to work with
04:56
an index for that
04:58
exactly and and you want your geography
05:01
to be representative and
05:02
depe and you have to make sure that your
05:06
geography
05:07
is going to reflect the needs of the
05:09
local authorities or you know
05:11
you know who's going to use this index
05:14
so
05:15
that's really important great wonderful
05:19
thank you thank you
05:21
[Music]
05:23
so we we said a little bit uh
05:27
in the the previous talk about uh
05:30
network distances being kind of
05:31
fundamental uh an opinion for this index
05:35
and there were some variables that don't
05:38
lend themselves to network distances so
05:40
air pollution particularly and so we
05:43
ended up combining
05:45
[Music]
05:47
combining those two different types of
05:49
data
05:50
um we we're going to say a little bit
05:53
more about
05:55
uh routine electronic thing um
05:58
but were there any particular
06:01
issues or uh problems with combining
06:04
these two different types of data which
06:06
are
06:06
at first glance might be quite different
06:08
in terms of how they're structured
06:14
yes um yeah they're also quite a few
06:17
challenges and that's
06:19
why you need a multi-dimensional index
06:22
is because you give you the flexibility
06:25
to
06:26
um play around with different methods
06:30
and try to find an appropriate one so to
06:32
be able to create an index at the end um
06:35
so the challenges were mainly the
06:37
challenges that everybody needs to
06:40
that is going to face everybody that
06:42
needs to
06:43
um create an index and that's the
06:46
distribution of each
06:48
variable so the distribution is not the
06:51
same for every single variable so you
06:53
have
06:54
for example their pollution is totally
06:56
different distributed
06:57
over you know england than for example
07:00
accessibility
07:02
to gp practices so
07:06
all these different distributions uh
07:09
to make sense we need to put them to
07:12
standardize them and put them in a
07:14
same let's say reigns so to be able to
07:17
combine them
07:19
so that's the biggest challenge that uh
07:21
researcher
07:23
faces uh for creating an index
07:28
wonderful great and we're going to pick
07:30
up on a few of those challenges later on
07:32
how we actually how you actually combine
07:34
those together
07:37
wonderful
07:40
so um you know this excuse me so this is
07:44
a kind of just a
07:45
a general overview of the index um
07:49
overall we've got better access to
07:51
healthy choices in
07:52
the periphery of the urban core and
07:55
worse access
07:56
and in urban core and rural areas
07:59
and we we touched previously on a bit of
08:01
the the kind of
08:02
urban rural divide and you know we could
08:05
do a whole another session on
08:07
urban rural sides of things uh so this
08:09
is just a kind of
08:10
a big picture for the data set itself
08:14
in terms of the you know that we've
08:17
found in this term benefit already
08:19
uh mdi multi-dimensional index and water
08:22
damage or indices
08:23
uh so overall it's a kind of framework
08:26
of methods
08:27
and uh could you give us a kind of
08:30
overview of
08:31
how how you use this for the our hr
08:33
index and we can pick up certain bits
08:36
later on
08:37
uh yes of course so the
08:39
multi-dimensional index
08:41
uh generally is uh has used from
08:45
many governments to create industries
08:48
all over the world
08:50
it's very standard framework to work
08:53
with and produce an index that
08:55
can be comparable and can keep track of
08:59
a phenomenon in your society
09:01
that you want to have a
09:04
picture and over time if you like but
09:08
also if you want to compare it with
09:09
other countries and so on
09:11
so the
09:15
imd index uh
09:20
sorry the multi-dimensional
09:24
index framework is actually
09:28
helps you to combine many different
09:30
indicators
09:31
and create one single
09:36
measure that is really simple
09:40
uh to be understood and
09:44
the most important to highlight the
09:46
phenomenon
09:48
in its uh area um
09:52
but that that can be a country that can
09:54
be a much more uh
09:56
targeted uh region and so on
09:59
okay great so it's um it's this kind of
10:02
comparable nature
10:04
that's key and it's it's almost though
10:07
it's a
10:09
a kind of very repeatable framework so
10:11
it's a kind of nice clear method that
10:13
can be applied in many different
10:15
countries if you know a different
10:17
country wanted to create its own
10:18
uh yes yes of course yes
10:22
that's the idea behind all the indices
10:24
that they they
10:26
for the creating a simple measure that
10:28
can be
10:29
repeated from you know other countries
10:33
or regions or
10:34
depending on who is going to be the
10:36
organization that
10:38
is going to use it and produce a really
10:42
comparable measurement right
10:45
wonderful so um
10:49
we said a bit about um why this method
10:53
um because of the comparability is there
10:56
is there any
10:56
were there any other methods you
10:58
considered and decided not to use
11:00
or anything more you want to say about
11:02
why you chose mdi
11:05
yeah you can have clustering
11:10
methods and
11:10
[Music]
11:13
also you can have a most more recent
11:16
time
11:18
methods like uh
11:21
random forest methodology to create a
11:25
kind of index from that if you like but
11:28
the idea here is that um
11:30
all these methods they have some
11:33
everything is driven by the data
11:35
and for example for clustering
11:39
the data sets are quite um
11:42
they need to be standardized properly
11:44
otherwise
11:45
your clustering is going to fail
11:49
so as we mentioned earlier our
11:52
our indicators are acquired you know the
11:54
distributions were
11:56
quite different so the clustering is
11:58
very challenging to be
12:00
achieved in this case so
12:04
the multi-dimensional indices they have
12:07
the advantage that they can
12:09
actually tackle these issues by removing
12:12
information
12:14
but aiming for one specific
12:18
target which is to create a single
12:20
measurement as in
12:22
a single index that can be interpreted
12:25
very clear and easily
12:29
so i think that's the main idea of
12:31
selection of
12:33
a multi-dimensional index for aha
12:37
great wonderful okay
12:41
so moving on to the next section
12:49
so um we've touched on a little bit
12:52
uh that the the multi-dimensional
12:54
indices approach
12:56
and framework is is the same one as used
12:59
by the index of multiple deprivation
13:01
which is a in the uk is quite a widely
13:04
known measure of deprivation and it's
13:06
used by
13:07
a local and central government for a
13:09
whole host of things
13:11
and can we can we say a little bit about
13:14
is that you did a good thing
13:15
but if you're using a similar method to
13:17
that or is it a bad thing or is there a
13:20
kind of
13:20
a bit of both in there i think it's um
13:24
i mean the reason we picked and we
13:25
followed the imd approach is
13:27
given that the ind is such a um
13:30
well used and well established
13:34
product data product it made sense to
13:37
try and follow
13:38
what they've done before rather than
13:40
trying to reinvent the wheel here
13:42
now that's not to say the imds approach
13:44
is the best
13:46
or the most appropriate way for building
13:47
indices but we we decided that given
13:50
that we was going to be
13:50
working with local governments it made
13:52
sense to try and justify it
13:54
by following their approach now
13:57
there's aspects of that approach which
14:01
have certain problems so like the imd
14:03
derives its weightings
14:05
for certain domains through more of a
14:07
data driven approach
14:09
and which we discussed in the first
14:11
videos might be necessarily not always
14:14
great um but i don't think there is
14:18
necessarily any right way and
14:20
you know developing indices it's more an
14:23
art than the science
14:24
there's lots of different ways you can
14:26
approach it there
14:27
a lot of them are appropriate and it's
14:29
about justifying your decisions
14:31
and like we discussed in the first video
14:34
we've made this all openly available
14:36
so it's up to people if they want to
14:38
reinvent our index and say actually we
14:40
think this is a much better way
14:41
of doing it and you know we encourage
14:43
that we'd love to hear from people who
14:45
think they've got a better idea and how
14:47
to improve it because
14:48
and you know i think that's the sort of
14:50
thing we can learn from
14:52
and the sort of thing that open
14:54
principles and science really do
14:56
help with great
14:59
okay thank you nice it's great to get
15:02
the background and and you know we've
15:05
got to remember there are there are no
15:07
absolutes it's
15:08
you know there's lots of different ways
15:09
of doing things that it's why
15:11
why you picked a certain method and how
15:13
you justify it because often the key bit
15:20
so um yes so we
15:24
we have a kind of overall uh view of
15:27
the the process um and one thing we've
15:30
alluded to a couple of times
15:31
but not really said much about is the
15:33
kind of first step so it's this
15:35
theoretical basis
15:36
and this is as i understand it is key to
15:38
building
15:39
any type of indexes well what are we
15:42
measuring
15:43
and and how do we want to measure it so
15:46
can we could you say a little bit about
15:48
the theoretical basis
15:50
and so what those kind of first steps
15:51
were and how they informed the latest
15:53
steps
15:55
so i think the the important step is
15:57
having a defined a very clear defined
16:00
index that you want to measure and
16:02
you've got perfect
16:04
all in good inputs to measure that
16:07
because one of the issues of building
16:08
any index is you'll get an answer either
16:10
way and you'll get a
16:11
metric that might be misleading or might
16:14
not have any
16:15
interpretable value um to the rest of
16:18
the world so i think spending a lot of
16:19
time
16:20
in this this stage is is the most
16:22
important part of building an index
16:24
and we in the um the kind of a heart
16:26
team spend quite a lot of time just
16:28
charting out what we should be doing
16:31
what we can measure what's useful
16:33
and that comes from talking to
16:35
stakeholder groups as well
16:37
and but you know the end of the day even
16:39
something that's very narrowly defined
16:41
so if we take the imd example they're
16:43
trying to measure deprivation but
16:45
there isn't actually a thing as
16:47
deprivation declaration is more a
16:49
concept that you're trying to measure as
16:51
a proxy through
16:53
a series of indicators it's the same for
16:55
a heart we're not necessarily measuring
16:57
anything other than we're trying to get
16:59
a handle on how healthy or unhealthy
17:01
environments are and here's a range of
17:04
data that might summarize
17:06
these issues and again it's about being
17:09
open and honest with people about
17:12
kind of what your your metric is about
17:14
and what it can and can't do i think is
17:16
important because
17:17
metrics get used and they get abused a
17:20
lot
17:21
which is one problem with developing
17:23
these things
17:25
okay great that's wonderful
17:29
and so we've got a kind of nice flow
17:32
chart
17:32
of the the process here um
17:36
so it costs us i was wondering if you
17:38
could take us through this kind of
17:39
process
17:40
we've got a few slides later on going
17:42
into more detail
17:44
uh but if you take us through a kind of
17:45
top level that would be really useful
17:47
and of course we've got to remember our
17:49
theoretical input is above
17:51
the the flowchart so it's kind of above
17:53
that of this graphic
17:55
yes of course and i think that there is
17:58
also another
17:59
level which is quite crucial is to
18:01
prepare the data
18:03
before you reach at the point to
18:04
aggregate them at the lsa level and
18:07
that's a really big
18:08
hurdle because um most of the data that
18:11
come
18:12
in different forms and you try to put
18:14
them
18:15
together and you have to check
18:18
for example your indicator if you have
18:21
the right
18:22
direction in terms of what you're
18:24
measuring
18:26
because you you know you're creating an
18:28
index and you want to help them
18:30
all your indicators looking to the right
18:33
to the same
18:34
uh direction instead of the one
18:37
canceling
18:37
the other one so
18:40
yeah so having you know all the cleaning
18:43
and the preparation of the data
18:45
then the first step of course is to
18:47
aggregate data at
18:48
lso level so that's the
18:51
level that we selected for our index to
18:54
be
18:56
you know the the base
18:59
and um then you have to
19:02
make a decision of the standardization
19:05
method
19:05
which can can be you know you can there
19:08
are many
19:09
methods around how to do that the
19:12
easiest method is the
19:15
ranking so you just create ranks
19:18
for each of your variables and then you
19:21
create a rank-based
19:22
normal transformation and this one
19:26
gives you the ability to emerge these
19:29
indicators together
19:30
there are other methods
19:34
like z scores which is the most common
19:36
one
19:38
but as i said before the set scores for
19:40
example can be very sensitive on the
19:42
distributions
19:44
of your data and
19:47
so that's why we selected ranking
19:50
because we have our data sets were quite
19:52
you know the distribution were quite
19:54
different uh between our domains
19:57
so yeah so for the first steps as i said
20:00
is to
20:02
run its indicator
20:05
then we use a rank based normal
20:08
transformation that's because we
20:10
when you want to merge you cannot merge
20:12
the ranking of
20:14
variables so what you want to do is to
20:16
create
20:18
a measure of a normalized distribution
20:21
that represents your ranking and
20:25
then uh if you require to apply some
20:29
waiting for these indicators then you
20:31
can do that
20:33
most of the times you can see
20:37
people using factor analysis to
20:41
determine these weights which is very
20:43
statistical
20:44
the statistical way to just look for
20:47
association
20:48
between your indicators uh but in our
20:51
case we
20:52
just found that there wasn't any need
20:56
for a factor analysis
20:59
weights to be introduced here so
21:02
we thought that the best approach the
21:05
best approach was to
21:06
just use equal weighted um
21:09
measures so then for
21:13
uh having beside this step then then
21:17
you build the four domains uh
21:22
and the next step again is
21:26
we need to put these for the domains
21:29
together
21:30
so that becomes again a challenge
21:34
because you have for example in how we
21:37
have two
21:39
domains that they are
21:43
uh promoting let's say the health
21:46
and and two um
21:51
and two domains that are actually
21:53
negating health
21:54
uh in our neighborhoods so
22:00
the two domains canceling no the one
22:03
cancels uh
22:04
each other yeah just just to recap
22:07
these are the four the two kind of
22:10
uh positive ones it was
22:14
hospital facilities and yes it's the
22:17
health domain
22:18
and the green and blue space yeah
22:21
so they're closer for example to a green
22:24
space that's a good thing so
22:26
the closer to be to the gp practice
22:28
that's a good thing
22:31
but the closer to you are in a
22:34
retail uh service outlet
22:37
then that's not very good and the same
22:41
happens if you have high numbers of
22:43
evolution okay
22:44
so having two and two domains
22:48
and what's going to happen if you just
22:50
equally weight them
22:51
then uh the
22:54
two domains they're going to cancel each
22:56
other
22:57
and for this reason we use exponential
23:01
transformation this one
23:03
this transformation helps us to avoid
23:06
this cancellation
23:07
and it's not something new that's uh
23:11
something that we borrow from imd and
23:14
it's well
23:15
explained in amv as well why
23:18
there is a need for art and we found
23:21
that
23:22
yes it is really important to apply the
23:24
exponential transformation and
23:26
avoid this type of cancellation and
23:29
it works very well and then again
23:33
we have we have to make a decision if we
23:35
want to
23:36
weight our domains in our case again we
23:39
have
23:40
two domains that are
23:43
good for health and to
23:46
negate health so we thought that
23:50
again we have to keep it uh
23:53
equal weighted uh for each of the
23:55
domains and then this
23:57
remates together and we create the high
23:59
index so that's
24:01
pretty much the methodology we follow
24:06
wonderful thank you costas it's great to
24:10
go through
24:11
those six stages and we're gonna uh
24:14
pick out a few more details about a
24:15
couple of them now
24:17
so we talked about the the
24:20
transformations
24:21
so this was in terms of when we go from
24:24
the kind of
24:25
aggregate data at the lso a level um
24:28
to do the the rank based normal
24:30
transformation and
24:32
can you say a little bit more about
24:36
how you picked which transformation you
24:38
did
24:40
yeah in the first stage let's say when
24:44
first of all the transformation the we
24:46
use transformation because we need to
24:48
merge
24:49
the indicators or the domains we need to
24:52
do
24:52
you know to merge in other measures if
24:55
it's possible to avoid the
24:57
transformation
24:59
then it's advisable to not do it so
25:02
it's really important the transformation
25:05
to be the last result
25:07
of your framework of your methodology
25:12
because every transformation introduce
25:16
uh errors mx alters your original data
25:20
all your original measures so in our
25:23
first
25:24
transformation the ranked based
25:27
normal transformation that's
25:30
really important to do this because we
25:33
took the decision
25:35
to move from the measures to
25:38
ranks so making this decision
25:42
to move to uh ranks then
25:45
there is no way to to combine
25:49
uh ranks uh between indicators
25:53
uh and the easy the easiest way
25:57
is to just uh introduce a rank based um
26:01
transformation normal transformation and
26:04
from this point
26:05
then actually each of your indicators
26:08
become again a measure
26:10
and that can which is standardized and
26:13
you can combine it
26:14
and make it one single domain
26:18
for our domain as i said before the
26:20
reasoning behind that was to avoid
26:22
cancellation
26:23
so that's why i use
26:26
the exponential transformation wonderful
26:29
great and and then the other bit
26:33
i i think you'll be used to play a
26:35
little bit about was the
26:37
the name weighting so once we
26:40
you have the the four domains and we
26:43
need to combine them
26:44
into one index and there's a few
26:48
different approaches
26:49
uh that you could take to this if she
26:50
takes through that
26:53
so the um the weightings it happens at
26:56
both the in
26:56
the inputs so our individual variables
26:59
whether that's green space whether it's
27:00
access to gps
27:02
we treat them all equally as well as the
27:04
domain scores which were all treated
27:06
equally the reason as i've said before
27:08
we did that is because we didn't have
27:10
any prior assumptions of what was more
27:12
or less important
27:14
so we just felt like an even waiting
27:16
just treating everything the same was
27:17
the most
27:18
open and fairest we could justify
27:20
essentially
27:21
now we've been pretty open about we
27:24
don't know if that is the best approach
27:26
but this is the approach that we think
27:29
works fairest
27:31
um but i think if you're developing your
27:33
own index you might want to consider you
27:35
know more of a theoretically ground
27:37
approach that you can justify different
27:39
weightings and you can actually back up
27:41
that evidence
27:42
with yes we think this domain or these
27:45
indicators are much more important
27:48
because at the end of the day your
27:49
weightings will be partly
27:51
drive your results of your index and if
27:54
something's
27:55
more weighted or less weighted then it's
27:58
going to affect how it contributes that
27:59
overall indicator
28:01
and so it's an important decision that
28:03
requires a lot of time and attention
28:06
the um imd they used the factory
28:08
analysis to kind of produce a more
28:10
data-driven approach
28:12
and but i think some of their
28:13
data-driven approach was still checked
28:15
over by a stakeholder group to say
28:17
does it make sense if we are wait these
28:20
domains versus the other domains but
28:23
you know whatever decision you make it's
28:25
hard to justify it and it comes back to
28:27
again that you you're measuring this
28:29
concept that doesn't necessarily exist
28:31
even if
28:32
it's healthy environments or deprivation
28:33
you're just trying to get a best
28:35
estimate of it
28:37
and and because of that it's then hard
28:39
to then
28:40
come up with a perfect solution to
28:42
measuring these things and you know it
28:44
is
28:45
it's an art in itself and how you graph
28:47
the art to get
28:48
a good representation of the world is is
28:51
tricky sometimes there's not necessarily
28:52
a set of steps you can follow
28:56
yeah no that's really useful
29:00
and then you know
29:03
with a lot of the kind of very data
29:05
science
29:06
heavy analysis that we see in other
29:09
subjects elsewhere
29:10
sometimes it's quite easy to forget the
29:13
the theoretical basis and the the real
29:15
world element
29:16
and but i think that's something that
29:18
you know as geographers
29:20
we know that's key key and crucial to
29:22
some of the work we do
29:24
and it is great to see how that's uh
29:26
influenced your your process here
29:31
so um the kind of penultimate area
29:35
if we um we will have some people
29:38
watching who are interested in creating
29:40
their own multi-dimensional indices
29:43
um either customizing the aha one or
29:46
creating their own one from scratch
29:48
we've covered a few key points already
29:50
in terms of the theoretical
29:52
justification for
29:53
what your index is doing in the first
29:55
place and then
29:58
what data goes into it so both in terms
30:00
of
30:01
what data is available and what's
30:03
appropriate um
30:05
and then the the balance of the domain
30:08
so
30:08
there's some more important than others
30:10
and are there any other
30:12
aspects that people should really bear
30:14
in mind if they're trying to undertake
30:16
their work
30:17
this work themselves i think
30:20
what is really important if you want to
30:23
get involved in creating uh
30:25
indices uh composite indices
30:28
uh is to uh go and read and see
30:33
other works other indices how they've
30:35
created them
30:36
because uh with the multi-dimensional
30:40
indices the knowledge comes from the
30:43
experience how how many of
30:46
uh how many how much experience do you
30:49
have on the field
30:51
and the best experience comes from
30:53
looking the work from others
30:55
how they tackle specific problems why
30:57
they did the specific use the specific
30:59
method so when we start ourselves
31:03
we have some you know knowledge we have
31:06
the skills but we didn't
31:08
we had to go and look and read about
31:11
uh you know the challenges the others
31:13
face
31:14
so to be aware of that and be able to
31:16
you know make the right judgment on
31:18
in terms of the selection of the method
31:20
and so on
31:21
so i think that the best way to start is
31:24
to just
31:26
go and look other indicators how they
31:28
built them and
31:30
why
31:33
yeah i decode those the comments of
31:35
costa
31:36
and also probably frame that you know
31:38
talking to people
31:39
talking to stakeholders talking to
31:41
people who might be users and getting
31:43
their views and
31:44
what they'd like to see in there and
31:46
building that kind of into your process
31:48
whether it's
31:49
building that into your theoretical
31:50
grounding or building it into what
31:51
customers are saying and the kind of
31:53
discussion of how other people have
31:54
built these things in the past
31:56
and because we've learned so much from
31:59
just working and talking to others
32:02
that we never think about and because
32:04
you're always coming from your own
32:05
particular point of view
32:08
and that's one of the great benefits of
32:11
working with a
32:12
interdisciplinary team getting people at
32:15
different people's opinions and
32:16
and talking to yourselves uh about what
32:20
you've done and
32:20
kind of building on that experience and
32:24
so in the in the future we will have
32:26
some interactive sessions with mark and
32:28
costas so if you do have questions do
32:30
come along to those
32:32
and we're happy to help out invariant if
32:34
we can
32:35
and but if you can actually talk to us
32:38
um
32:38
please do and we're very happy to help
32:41
that
32:47
so we've we kind of covered the the kind
32:50
of final point already really
32:51
um in terms of the kind of strengths and
32:54
weaknesses of the
32:56
multi-multi-dimensional
32:57
indices uh framework um you know and
33:00
there's
33:00
lots and lots of different elements uh
33:03
for it and there's a lot of pros and
33:04
cons to each
33:05
decision you make um is there anything
33:08
else you'd like to add
33:10
to that side of things
33:15
i think uh one thing that i would have
33:18
to mention is that the strength of the
33:19
weaknesses of its
33:21
ear index depends on the
33:24
methods used in this
33:29
index so for example in our case we have
33:34
a user ranking method that means that
33:37
we have lost any information related on
33:42
the
33:42
range of uh we cannot make div we cannot
33:45
measure the difference between
33:46
areas but we can know for example if one
33:50
area is worse than another but we don't
33:53
know
33:54
we don't know the degree of this change
33:57
um but other indices if for example you
34:02
they use z scores then they keep this
34:04
information and then they can
34:06
answer this question so it's very it's
34:09
really important to
34:12
when you see an index to see which
34:15
what is the you know the weaknesses of
34:18
this index but also what is the
34:20
advantages of this index
34:23
so that's really important
34:27
to be aware of these limitations
34:31
so in terms of the r index so we could
34:35
to give an example we could say that an
34:37
area of
34:39
20 is um less surprising an area of 24
34:44
but it's we can't say it's half the
34:46
decoration the area of ten
34:49
yes yeah that's right okay so it's the
34:52
kind of
34:54
uh the score at the end and how how we
34:56
help people interpret that
34:59
and that's the same for the imd as well
35:01
the
35:02
because it's pretty much similar
35:05
methodology
35:06
what we have to understand here is that
35:10
why these indices have been created the
35:12
way that they have been created
35:14
the index the indices have created this
35:18
way because what we
35:19
want to do is to highlight areas in need
35:23
so in our cases with body health
35:27
accessibility or in imds with uh
35:32
in terms of deprivation but the aim is
35:35
to highlight these areas
35:37
then the uh
35:40
the planners they're gonna go to these
35:42
areas and do
35:44
extensive analysis on these areas and do
35:46
you know
35:48
collect more information and try to
35:50
understand better what's happening in
35:52
this area why
35:53
we have this area highlighted
35:56
and so on so it's an indicator again
35:58
it's the name index
36:00
doesn't give you all the answers
36:05
great thank you for that and
36:09
mark any any kind of parting words of
36:12
wisdom
36:14
uh i think costas has summed up
36:15
brilliantly i think the the one thing we
36:17
haven't discussed about is thinking
36:19
about how you release your index
36:20
and actually does whether you're going
36:23
to go for a fully open approach or a
36:24
closed approach but also
36:26
really building in tools thinking about
36:28
how your users might use these
36:31
products so we have the the cdrc maps
36:33
platform which is really good for
36:35
having people to interact with our
36:37
resource but supporting that with
36:39
appropriate metadata
36:40
resources guides it's a lot of work and
36:43
it shouldn't just be we've released this
36:45
index it should be we've released this
36:47
index and
36:48
provided a lot of support to help and
36:51
maximize the reach
36:53
of how people might use and engage with
36:55
it
36:58
right now that that's a really really
37:00
really good point
37:01
and it's something that we should all
37:04
remember
37:05
the the work we're doing how to how do
37:07
people use it and how can you support
37:09
that and
37:10
you know we hope the this set of videos
37:12
has been
37:13
uh useful for for people using the index
37:16
and
37:17
thinking about creating or editing their
37:19
own and please do
37:21
let us know how you get on please do
37:23
provide some feedback
37:24
in the comments or or send us an email
37:27
there's uh
37:28
links below the video and there's also a
37:31
workbook
37:32
associated with this session where we
37:33
take you through the process in our
37:36
creating the multi-dimensional index we
37:39
use our heart as an example we've
37:41
simplified a significant chunk of the
37:42
process but it will give you a good
37:44
overview and there's various points
37:47
where we
37:48
explain decisions that have been made
37:50
for our heart
37:52
and and if you're applying that to your
37:54
own data do remember
37:55
you know the key points that we've
37:58
talked i don't think we can
37:59
emphasize that enough for it
38:04
so thank you very much everyone for
38:07
listening and thank you
38:08
mark and costas for your input it's
38:10
great to have a discussion
38:12
and uh hear about your your experiences
38:15
and uh i look forward to the next time
38:17
we all meet
38:19
thank you very much thank you