Statistical predictions for numbers of CPP and LAC for the West Midlands Local Authorities, based on their level of deprivation
Statistical predictions for numbers of CPP and LAC for
the West Midlands Local Authorities, based on their
level of deprivation
Darrell Harman & Chandan Kaur, November 2013
Why this study was carried:
Directors of Children‟s Services have a statutory duty to
„address‟ needs within their areas.
“The DCS is responsible for securing the provision of services which
address the needs of all children and young people, including the
most disadvantaged and vulnerable, and their families and carers.”
“The LMCS is responsible for ensuring that the needs of all children
and young people, including the most disadvantaged and vulnerable,
and their families and carers, are addressed.”
Source: Statutory guidance on the roles and responsibilities of the
Director of Children’s Services and the Lead Member for
Children’s Services. April 2013
Impact of the economic recession
The economic recession has caused a situation where
need (deprivation) has increased since 2008
At same time, as a result of the Coalition Government‟s
Deficit Programme, Council budgets are reducing
There is an even greater need to understand how
increased need translates into demand for social care
Results may help Children‟s Services to understand if:
o their level of demand = predicted demand
o they are overspending or underfunded?
Extensive research on the relationship between
deprivation and social care demand
(several cited in Jan 2011 report)
NSPCC Child protection research briefing between Poverty
and child maltreatment (2008) – research evidence showed
there is an association between poverty and an increased
risk of child maltreatment, particularly neglect and physical
Sidebotham, P. and Heron, J. (2006): Child maltreatment in
the “children of the nineties”. A cohort study of risk factors found that the biggest risk factor to CPP listing was Parental
Poverty. (10:1 odds ratio, domestic abuse 2:1)
Bywaters, P. (2013) „ Inequalities in Child Welfare‟ – clear
relationship between the child‟s chances of being looked
after or being subject to a CPP with the overall deprivation
in the LA in which they lived.
Why JSA and not English Indices of
IDACI, IMD (2010) is actually based on 2008 data – so takes no
account of the worst depression since the 1930s.....
JSA counts are available for each month, a month later and at a
small area level, more real time and released in a timely fashion.
JSA is a good proxy of need as it is part of IMD and IDACI
West Midlands as a region - context
rate in Great
Total "CPP and LAC Population Rate" and JSA rates for the West Midlands Region
P/t jobs and
effect and LoS
Baby P case: high
national media profile
Mar 09: Publication of
Lord Laming Report
Apr 08: Publlic
Total CPP+LAC Rate
West Midlands JSA Rate
CPP/LAC increased well before Baby P impact and
soon after recession commenced.
Total LAC and CPP Rate per 10,000 under 18 population
This statistical model is based on
the real system and flows
The above system can be mathematically solved by a series
of difference equations if all counts/rates are known:
e.g. if t= the count/rate at the end of month or year
and ∆t = the count at the end of the following month or
CPP(t+ Δt) - CPP(t) = (CiNOUT to CPPIN)Δt + CPPINΔt CPPOUTΔt – (CPPout to LACIN)Δt LAC(t+ Δt) – LAC(t) =
LACINΔt - LACOUTΔt + (CPPout to LACIN)Δt
Characteristics of a useful predictive model
i) Firstly the predictor variables have to be statistically significant with the
variable that is being predicted i.e. in this case, CPP and LAC
ii) Secondly the model has to have high predictive power i.e. R-squared is
(iii) The 95% confidence limits of the mean predicted variable are narrow
enough to be useful and provide LAC ranges so more realistic for
Our statistical model meets these three tests= useful predictive model!
What does statistically significant mean? Data showed that there is a relationship
between the predictor(s) and predicted variable, which could not be explained by
High R-squared = the model formulated with predictor variables can explain a large
degree of the variation in LAC or CPP.
• It is important to remember that a high correlation
coefficient or a high R-sq does not prove causation. In
other words, a strong correlation with Unemployment rates
does not mean that if parents become unemployed that
their children will require support from Children‟s Social
Care Services; it only applies at population levels.
• One explanation could be that unemployment can further
increase the risk of pre-existing family dysfunction /
stresses (e.g. family breakdown, domestic violence)
eventually leading to a crisis. Research supports this
Model 1: National I52 councils
Multiple regression model developed based on snap-shot data (31st
March 2013) from 152 LAs in England.
Multiple regression model = more than one predictor variable used
JSA rates 31st March 2013, Proportion of under 5s within under 18s
population, Proportion of Asians within local population (negative
Snap-shot data taken from DfE 31st March 2013 statutory returns, 2011
Census and www.nomisweb.co.uk for JSA rates.
Outcome: Predictions for LAC populations were the most useful
from this method.
Model 1 LAC mean predictions with 95% CI from
Multiple regression model
* LA has UASC
p/t work and
need is likely
to be closer
JSA based model could be underestimating risk…
Due to economic structural changes (increased underemployment and p/t work) and benefit sanctions;
The combined affects of all 2010-2013 reforms excluding Universal Credit leave Lone Parents
worse off by £32.67pw (7.8%), and couples with children by £43.86 (5.6%)
1 in 10 workers is „under employed‟, increasing by 42% across all occupations in four years to
E.g. Between October 2013 and June 2013, 7,590 JSA claimants in Walsall had their
Data match CiN with Benefit cap data and target support
It maybe that JSA is becoming a less accurate predictor of
need, so make use of the upper 95% confidence intervals
Child Poverty Strategy – a corporate response is needed
to mitigate LAC costs pressures on council.
Model 1 useful for predicting LAC only
Model 1 had low predictive power for predicting
LAC admissions and CPP listings.
This was similar to what was found when using IDACI
Poor predictive power due to poor correlation.
Why would this happen, compared to LAC population high
Preventative work more effective at lower levels of risk
LAC admissions volatile (large sibling groups)
Risk appetite varies by councils (see LAC/CPP ratio variance see next slide)
Throughput of CPP is higher than LAC.
Ratio of LAC to CPP by council over time
LAC:CPP ratio, 31st March
Herefordshire, County of
Telford and W rekin
Average LAC:CPP ratio
WMids LAC + CPP counts
Are LA risk thresholds influenced by volumes of LAC and CPP? As volumes increased across
region, the ratio has reduced – is this related to budget pressures?
LA variation (31st March 2013) from West Midlands average regional
LAC to CPP ratio
Lower risk threshold - "Risk adverse"
Higher risk threshold - "Manage Risk"
Model 2: West Mids LAs only, over time
Simple linear regression models using 14 West Midlands LAs data
over time (end of year CPP and LAC rates vs JSA rates from
2001 to 2013). For example:
Due to the large variation in JSA rates; 14 LA‟s
allocated into either a „Higher Deprivation‟ band or
„Lower Deprivation‟ band.
Outcome: predictions for LA‟s good for;
and CPP populations including CPP listings
and LAC admissions.
and Predictive power ranged from 47% to 87%!
Predictions for LAC from Models 1 and 2
Blue: >>10% difference outside range on both models
Yellow: outside ranges but <=10% difference outside range on at
least one model
Green: inside range on at least one modelwww.walsall.gov.uk
If your council is outside of predicted ranges on
You may plan to change your LAC and/or CPP numbers.
You may wish to explain them based on local evidence;
(i) Technical – wrong coding in statistical returns
(ii) Interpretational – e.g. of indicators
(iii) Situational e.g. large number of UASC or higher level of drug misuse
(iv) Operational - were thought to have the most influence and was
seen to be more in local authority control:
How children‟s services were resourced and delivered,
the extent of inter-agency working,
professional expertise and management support
and the availability of family support services.
Oliver et al (2001) reasons for variance in CPP or LAC numbers not
accounted for by deprivation:
Evidence from London
Children looked after rate, per 10,000 children aged under 18
Findings from a London Study, Feb 2013:
•Proactive approaches to reduction usually work
•Focus on flows (especially early discharges)
•Lower caseloads are linked to lower LAC; av. for inner London
was 17 cases.
•Spend needs to be sufficient and focussed
•Early help is as yet unproven
•Senior managers commented on a direct correlation with rise
of LAC numbers following either a poor inspection or a high
profile Serious Case Review.
Statistical models can provide estimates for predicted
demand based on current levels of deprivation.
JSA is a good proxy measure of deprivation –
captures impact of economic recession more fully.
Two statistically significant models developed predict
LAC and CPP
– Model 1: National snap shot data – predicts LAC
– Model 2: Regional data – predicts LAC but CPP
populations, CPP listings and LAC admissions as
Structural economic change and benefit capping/sanctioning
means demand more likely to be nearer upper 95% CI of
Models 1 and 2.
If your council is outside these predicted ranges then you
may wish to
Understand and explain variance due to local factors. Perform a
systems „health check‟
- Increase/reduce LAC and Examine LAC/CPP ratio
- Ensure a corporate Child Poverty reduction targets targets CiN