Investing in Prevention & Support

Spending on family support, children’s centres, young peoples’ services, and other forms of help and child welfare interventions in England, 2009-10 to 2021-22

Dr. Calum Webb
Sheffield Methods Institute, the University of Sheffield
c.j.webb@sheffield.ac.uk

Postdoctoral Fellowship
PF21\210024

Cite as: Webb, Calum (2024). Investing in Prevention & Support: Spending on family support, children’s centres, young peoples’ services, and other forms of help and child welfare interventions in England, 2009-10 to 2021-22. The University of Sheffield. Presentation. https://doi.org/10.15131/shef.data.26348338

What is effectiveness?

Throughout this report I use the term "effectiveness" to refer to the extent that spending on preventative services reduces rates of child welfare interventions.

However, we should not lose sight of the fact that this is only one part of what effectiveness means.

Effectiveness also means the extent to which services hear parent's, children's, and other family members' voices; respond to community needs; adhere to process; champion children's and human rights; challenge inequality and oppression, and many other things.

Please keep this in mind when considering the findings and language.

Some key terms

All rates refer to average rates at 31st March of each financial year.

Children looked after rate refers to the rate of children in care per 10,000 child population, unless otherwise specified. This could be care in a foster home, residential home, or in supported independent accommodation.

Child protection plan rate refers to the rate of children on a child protection plan per 10,000 child population. A child protection plan usually follows a section 47 enquiry where a safeguarding assessment by a local authority social worker or social work team finds that a child requires a plan according to the statutory guidance.

Child in need rate refers to the rate of children with 'child in need status' under Section 17 of the Children Act 1989 per 10,000 child population. This section of the Children Act places a duty on local authorities to provide additional support to children who are unlikely to achieve or maintain a reasonable standard of health or development, or whose health or development is likely the be impaired, without the provision of additional services, or, if the child has a disability.

Preventative services spending refers to spending on children's centres and children under 5; family support services; services for young people (including youth work, youth centres; substance misuse and teenage pregnancy services); and 'other' spending not related to children looked after, safeguarding, or youth justice.

Methods & Data (The Short Version)

148 Local Authorities (exl. City of London, Isle of Wight, Isles of Scilly, re-unified Bournemouth & Poole)
13 time points (2009-10 to 2021-22) without time-varying child poverty rate; 8 time points with time-varying child poverty rate (2014-15 to 2021-22)
Data from Department for Education (Child Welfare Intervention Rates, Section 251 Returns); Office for National Statistics (Population Estimates); Department for Work & Pensions (Children in Low Income Households)
Spending figures adjusted for Service Providers Inflation Index (2022 prices)
Within-between multilevel models estimated throughout within a Bayesian framework, using R and Stan
- Corrects for poor comparability of Section 251 returns by focusing on variation within local authorities
Missing data imputed through Bayesian 2-level model based multiple imputation by chained equations. Most missing data for CLA counts for Black (27%), Asian (32%), and 'other' ethnic group (41.5%) due to censoring
All code and data made available through open repositories upon publication

I.I Findings: The evidence strongly suggests that investing in preventative services reduces rates of children looked after and children on child protection plans.

Does spending on preventative services reduce child welfare intervention rates?

Each £100 decrease in family support, youth services, children's centres, and other preventative services per child was associated with an increase of around 3.5 children in care per 10,000 over the following two years...

And around a 2 per 10,000 increase in the number of children placed on child protection plans (between 2009-10 to 2021-22).

Decreased spending in a given year was associated with higher children in need rates the following year, but lower rates in the same year between 2009-10 and 2021-22.

per 10,000

Does spending on preventative services reduce child welfare intervention rates?

This would equate to around 20 more children in care in an average sized local authority...

And around 10 more children children placed on child protection plans (between 2009-10 to 2021-22), on any given day of the year.

in an average local authority

Average reductions in spending on preventative services between 2009-10 and 2021-22 can explain approximately 60% of the average increase in children looked after rates within local authorities.

Between 2014-15 and 2021-22 reductions in spending can explain around one third of the increase in children looked after rates, with increases in child poverty able to explain nearly the entire remainder of the observed increase.

Between 2014-15 and 2021-22...

It is possible to include variation in rates of children living in low income families within local authorities over time.

Every £100 per child spent on preventative services was associated with a 0.18 percentage point decrease in children in low income families in the same year, and a 1.38 percentage point decrease in children in low income families in the following year.

The median local authority saw a 4.1%p increase in children in low income families between 2014-15 and 2021-22; around one fourth of this increase can be attributed to cuts in services to support families with children.

>2/3

of the effectiveness of spending was attributable to decreases in child poverty

Between 2014-15 and 2021-22...

The overall effect of spending on rates of children looked after remained broadly similar between 2014-15 and 2021-22, increasing slightly to a 3.86 per 10,000 decrease in children looked after rate for a £100 per child increase in spending within a local authority

A substantial amount of this was due to the effect of spending on reducing rates of children in poverty: 2.75 per 10,000 of the decrease

However, during this time period, increases in spending tended to be associated with increases in child protection plan rates.

The "cost" of investing in support

'Costs' associated with children being looked after are notoriously difficult to estimate. The moral and human costs of not supporting families should outweigh any economic costs.

A £100 per child increase in spending on preventative services would cost approximately £1.11billion in England.

Of this, we might expect around 30% of the costs to be returned through reductions of children in care (at £77,200 per child per year) and improved mental health (Ford, et al. 2007) and criminal justice outcomes (Leyland, et al. forthcoming) alone.

According to estimates by CPAG, we would expect reductions in child poverty to result in returns of approximately £3.05billion; 275% of the investment.

When spending on preventative services is low, more spending on safeguarding services tended to lead to increased rates of children in care

In years with average levels of preventative spending within local authorities, a £100 per child increase in safeguarding (child protection social work) spending was associated with a 3.69 per 10,000 increase in children looked after rate.

By contrast, in years where spending was £200 per child higher than average, a £100 increase in safeguarding spending per child would be predicted to be associated with a negligible 0.09 per 10,000 increase.

In 2009-10 and 2010-11, around 28 percent of local authorities had preventative services spending that was £200 or more higher per child than average between 2009-10 and 2021-22

There was also evidence that higher spending on preventative services could attenuate the relationship between increases in child poverty and children in need rates.

I.III Findings: Investing in preventative services was effective for reducing all forms of child welfare interventions before 2013, but only effective for reducing rates of children in care since 2017.

Before 2012, a £100 per child increase in preventative and supportive services within a local authority was associated with around a 2 to 2.2 per 10,000 decrease in children looked after rate; a 2.6 to 2.8 per 10,000 decrease in child protection plan rate; and 4.6 to 5.9 per 10,000 decrease in children in need rate. The effect was also more consistent across local authorities.

Between around 2013 and 2017, there was no strong evidence that increased spending was associated with decreased rates of intervention.

Between 2017 and 2022, spending was associated only with increasing effectiveness for reducing children looked after rates, from a 2.3 per 10,000 reduction to a 7.9 per 10,000 reduction per £100 per child.

I.IV Findings: Spending on preventative and supportive services does not have the same impact on rates of intervention everywhere.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Children Looked After rate

In 76.8% of local authorities, we would expect an increase in spending on preventative services to result in a decrease in children looked after rate over two years.

Of these, 60.3% had a probability of direction greater than 0.89, and 19% had a probability of direction between 0.75 and 0.89.

31.1% of local authorities had estimated reductions in CLA rates that were at least twice as large as the median estimated change.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Child Protection Plan rate

In 85.4% of local authorities, we would expect an increase in spending on preventative services to result in a decrease in child protection plan rate over two years.

However, of these, only around one third had a probability of direction greater than 0.89, with around another third having a probability of direction between 0.75 and 0.89.

11.9% of local authorities had estimated reductions in CPP rates that were at least twice as large as the median estimated change.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Children in Need rate

Relationships between spending and children in need rates were far more mixed, with 54.3% of local authorities estimated to see a decrease in children in need rate over two years for an increase in spending on preventative services.

However, of these, 40.2% had a probability of direction greater than 0.89, and 26.8% had a probability of direction between 0.75 and 0.89.

And around 43% of local authorities had estimated reductions in CIN rates that were at least twice as large as the median estimated change.

Children should be able to expect to be supported to the same extent, no matter where in the country they are, to live safely at home.

However, please keep in mind that just because spending on preventative services does not reduce intervention rates everywhere, does not mean that these local authorities don't excel in other important ways.

More effective spending means children's services are more dependent on these services being funded.

In the context of austerity, this means that many of these local authorities have more often than not seen dramatic increases in their intervention rates as their ability to fund preventative and supportive services has been diminished.

Local authorities do not tend to have spending effects that reduce all types of intervention rates to similar extents.

There are some exceptions, for example, LAs in the Tees Valley or the cluster of Hampshire, Portsmouth, and Southampton, as well as stand-alone local authorities like Cumbria, Sefton, and Bradford. But it is more common for local authorities to see spending reduce some forms of interventions but not others.

There was some tendency for preventative services to be effective for reducing either child protection plan rates, or children in need rates, perhaps reflecting differential capacities to respond to more upstream needs.

This variation can help inform local policy and practice learning.

Spending effectiveness was not strongly associated with general levels of child poverty.

There was moderate evidence that local authorities that had protected preventative spending had more effective services for reducing children looked after rates, on average.

There was a 93.4% probability that greater reductions in preventative spending between 2009-10 and 2021-22 were associated with less effectiveness per £1 per child invested in prevention.

There was also moderate evidence that local authorities that had higher overall spending had more effective services for reducing children in need rates.

There was a 93.9% probability that higher overall spending between 2009-10 and 2021-22 was associated with greater effectiveness per £1 per child invested in prevention.

I.V Findings: There were substantial inequalities in the effects of preventative spending on child welfare intervention rates across age, gender, and ethnicity.

Between 2009-10 and 2021-22, increases in spending were most strongly associated with decreased rates of intervention among children aged 16 and older, children aged 10 to 15, and children aged under 1.

Indicative changes in rates: -16.9 per 10,000 (16 and older); -4 per 10,000 (10 to 15); -1.7 per 10,000 (Under 1); +0.42 per 10,000 (5 to 9); 1.92 per 10,000 (1 to 4).

Between 2014-15 and 2021-22, the effect for one to four year olds was reversed, and the effect for children aged over 10 was far more muted, potentially due to the removal of the mediating factor of poverty.

The findings suggest that child poverty mediates a far greater extent of spending effectiveness for older children, and tends to be associated with much larger increases in rates of care for younger children.

The predicted effect of a £100 per child increase in preventative spending varied dramatically between local authorities.

The figures on the right-hand side represent the middle 50 per cent range of the expected effects across local authorities in England.

While increased spending on preventative and supportive services is very likely to decrease rates of intervention across almost all age groups in a statistically average local authority, this is not necessarily the case everywhere.

-11% to +4%
Under 1

-10% to +16%
1 to 4

-10% to +10%
5 to 9

-14% to +2%
10 to 15

-26% to -2%
16 and older

Local authorities tended to have a tendency towards spending that was either effective for reducing rates of care for younger children, or effective for reducing rates of care for older children.

One somewhat outlying correlation was between variation in the effectiveness for reducing rates of care for over 16 year olds and effectiveness for reducing rates of care for Under 1s, which may indicate the role that teenage pregnancy services have.

Between 2009-10 and 2021-22, the effect of increased spending on children looked after rates was notably stronger among the male child populations than the female child population.

Though the effect was substantial for both groups of children.

However between 2014-15 and 2021-22, the direct effect of spending on preventative and supportive services in isolation of any mediating effect of poverty was no longer substantial for female children.

But, female children's rates of care within local authorities over time were more strongly associated with increases in child poverty.

There was only slightly more variation across local authorities for male children than there was for female children.

Further, there was a strong correlation between the effect of spending on care rates for female child and the effect of spending on care rates male children, indicating that local authorities with effective spending for reducing rates of care among male children also tend to have effective spending for reducing rates of care among female children, and vice versa.

-14% to no change
for male children

-11% to +3%
for female children

0.793

Correlation between effect of spending for female children and effect of spending for male children across local authorities

The effect of spending was more strongly associated with changes in rates of children from minoritised ethnic groups in care.

For Black children, each £100 per child decrease in spending was associated with a 4.1 to 5.6 percent increase in rates of care (equivalent to a 4.5 per 10,000 increase in rates of children in care); a 5 percent increase for Asian children (1.42 per 10,000); a 6 percent increase for mixed heritage children (5.83 per 10,000); and a 21 percent increase for 'other' ethnic groups (-19.67 per 10,000). For White children, a £100 per child decrease was associated with a 2 percent increase (-1.34 per 10,000).

Between 2014-15 and 2021-22, after removing any mediating effect of poverty, we find that the direct effect of increased spending among most minoritised ethnic groups is no change in rates of care.

This implies that either a) preventative and supportive services have become less effective for minoritised ethnic populations over time, and/or b) the vast majority of the benefits of investing in support for these populations are mediated through the effect this spending has on child poverty.

Children from minoritised backgrounds faced far more spatial inequality in the effect of spending across local authorities than White children.

The predicted effect of a cut to, or increase in, preventative and supportive spending was more dispersed across local authorities amongst ethnic minority populations. Even if the average effect of investment in spending is more beneficial for minoritised populations, this was very spatially unequal.

3.5 times more variation
for Black children compared to White

3.5 times more variation
for Asian children compared to White

2.3 times more variation
for Mixed Heritage children compared to White

4.3 times more variation
for other ethnicity children compared to White

There was an alarming lack of correlation between the relationship between spending on preventative and supportive services and rates of care across different ethnic populations.

High correlation would indicate investment in preventative and supportive services, while variable, is approximately aligned across all ethnic groups. These relatively low correlations suggest that many local authorities have services that are associated with rates of children in care in very different ways depending on the ethnicity of the child population.

Despite the fact that we might expect the relationship between spending and rates of care to be stronger in urban local authorities, this did not seem to be the case.

The relationship was broadly similar across local authorities in urban, rural, and rural with significant urban regions, with the relationship implying slightly more effectiveness in areas with at least some rurality. This may reflect differences in access to or the capacity of services, or the way experiences of poverty differ in ways where services may be more effective than cash transfers.

I.VI Findings: Investing in Prevention & Support: Summary

There was strong evidence to support the idea that investing in prevention and support effectively reduces rates of child welfare interventions, specifically, rates of children in care or on child protection plans.

Reduced spending on preventative and supportive services can explain a substantial proportion of recent increased in rates of children in care, but increases in child poverty have been more impactful. However, investing in support can also assist with tackling child poverty, and both goals should be pursued in tandem.

Investment in support and prevention is very likely to be at least cost-neutral, and a cursory economic analysis using cost estimates from respected sources suggests investing in these services saves money.

Investing in only safeguarding services is likely to increase rates of children placed in care if services to support families and young people are not also adequately resourced.

However, we must be mindful of both inequalities over time, place, and between different groups. Investing in preventative services may be effective for reducing rates of intervention on average, but such investment needs to be inclusive, and should not depend on when a child is growing up, where they are growing up, or their age, gender, or ethnicity.

This complexity is difficult to manage, and more complexity in the form of intersectional analyses is needed, but mapping out variation across local authorities creates great capacity to learn from one another to address inequalities in the effects of support.

II.I Methods: If spending has such a large effect, why has this not been found in studies before?

The major reason why the findings from this study differ from those of other studies is that a within-between regression model is used, as well as the inclusion of a lagged term.

A groundbreaking study, by Dr. Davara Bennett et al.¹, used these models before to explore differences between one to four year olds and older children.

There are other differences discussed in the report, including the use of Bayesian statistics and multiple imputation. But the form of the model is the most important reason why the findings may differ.

Standard Regression Model

$\bar{Y} = B_0 + B'X'$

Random Intercepts Multilevel Regression Model

$\bar{Y_{ij}}=B_0+B'X'_{ij}+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Within-between Regression Model

$\bar{Y_{ij}}=B_0+B'_w(X'_{ij}-\bar{X}'_j)+B'_b\bar{X}'_j+U_{0j}\\U_{0j}=N(0,\sigma_0)$

¹ Bennett, D.L., Webb, C.J., Mason, K.E., et al., (2021). Funding for preventative children’s services and rates of children becoming looked after. Children and Youth Services Review, 131, p.106289.

Differences in the substantive interpretation of coefficients:

$B'_w$ — What is the association between changes in $X$ around the mean for groups $j$ and changes in $Y$ around the random intercepts estimated by $U_{0j}$ ?

$B'_b$ — What is the association between group mean differences in $X$ and group mean differences in $Y$ ?

In this context, the relationship between changes in spending/poverty within a local authority over time has with rates of intervention within that local authority over time, are separated from the relationship between local authorities average levels of spending/poverty and their average rates of intervention.

Standard Regression Model

$\bar{Y} = B_0 + B'X'$

Random Intercepts Multilevel Regression Model

$\bar{Y_{ij}}=B_0+B'X'_{ij}+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Within-between Regression Model

$\bar{Y_{ij}}=B_0+B'_w(X'_{ij}-\bar{X}'_j)+B'_b\bar{X}'_j+U_{0j}\\U_{0j}=N(0,\sigma_0)$

"The conflated slope in the [uncentered] UN Model lies between its within- and between-cluster effect, but is equal to neither. The conflated estimate is therefore meaningless and cannot be interpreted."

Yaremych, et al. 2021. Centering Categorical Predictors in Multilevel Models: Best Practices and Interpretation. Psychological Methods.

Why does this matter so much for studying spending on preventative and supportive services?

Lack of comparability of spending data between local authorities, but better comparability within local authorities.
Lack of comparability between usage of some interventions between local authorities, but better comparability within local authorities.
Fundamentally different causal directionality at the between level when compared to the within level.

Between local authorities

Why does this matter so much for studying spending on preventative and supportive services?

Lack of comparability of spending data between local authorities, but better comparability within local authorities.
Lack of comparability between usage of some interventions between local authorities, but better comparability within local authorities.
Fundamentally different causal directionality at the between level when compared to the within level.

Within local authorities

A visualisation of model complexity up to within-between regression

$\bar{Y}=B_0+B_1S$

A visualisation of model complexity up to within-between regression

$\bar{Y}=B_0+B_1S+U_{0j}\\U_{0j}=N(0,\sigma)$

A visualisation of model complexity up to within-between regression

$\bar{Y}=B_0+B_1S+B_2P+U_{0j}\\U_{0j}=N(0,\sigma)$

A visualisation of model complexity up to within-between regression

$\bar{Y_{ij}}=B_0+B'_{w1}(S'_{ij}-\bar{S}'_j)+B'_{b1}\bar{S}'_j+B'_{w2}(P'_{ij}-\bar{P}'_j)+B'_{b2}\bar{P}'_j+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Dr. Calum Webb

Sheffield Methods Institute
The University of Sheffield
The Wave, 2 Whitham Road
Sheffield
S10 2AH

c.j.webb@sheffield.ac.uk

Postdoctoral Fellowship
PF21\210024

Investing in Prevention & Support

Spending on family support, children’s centres, young peoples’ services, and other forms of help and child welfare interventions in England, 2009-10 to 2021-22

Dr. Calum Webb
Sheffield Methods Institute, the University of Sheffield
c.j.webb@sheffield.ac.uk

Postdoctoral Fellowship
PF21\210024

What is effectiveness?

Throughout this report I use the term "effectiveness" to refer to the extent that spending on preventative services reduces rates of child welfare interventions.

However, we should not lose sight of the fact that this is only one part of what effectiveness means.

Please keep this in mind when considering the findings and language.

Some key terms

All rates refer to average rates at 31st March of each financial year.

Methods & Data (The Short Version)

148 Local Authorities (exl. City of London, Isle of Wight, Isles of Scilly, re-unified Bournemouth & Poole)
13 time points (2009-10 to 2021-22) without time-varying child poverty rate; 8 time points with time-varying child poverty rate (2014-15 to 2021-22)
Data from Department for Education (Child Welfare Intervention Rates, Section 251 Returns); Office for National Statistics (Population Estimates); Department for Work & Pensions (Children in Low Income Households)
Spending figures adjusted for Service Providers Inflation Index (2022 prices)
Within-between multilevel models estimated throughout within a Bayesian framework, using R and Stan
- Corrects for poor comparability of Section 251 returns by focusing on variation within local authorities
Missing data imputed through Bayesian 2-level model based multiple imputation by chained equations. Most missing data for CLA counts for Black (27%), Asian (32%), and 'other' ethnic group (41.5%) due to censoring
All code and data made available through open repositories upon publication

I.I Findings: The evidence strongly suggests that investing in preventative services reduces rates of children looked after and children on child protection plans.

Does spending on preventative services reduce child welfare intervention rates?

And around a 2 per 10,000 increase in the number of children placed on child protection plans (between 2009-10 to 2021-22).

Decreased spending in a given year was associated with higher children in need rates the following year, but lower rates in the same year between 2009-10 and 2021-22.

per 10,000

Does spending on preventative services reduce child welfare intervention rates?

This would equate to around 20 more children in care in an average sized local authority...

And around 10 more children children placed on child protection plans (between 2009-10 to 2021-22), on any given day of the year.

in an average local authority

Average reductions in spending on preventative services between 2009-10 and 2021-22 can explain approximately 60% of the average increase in children looked after rates within local authorities.

Between 2014-15 and 2021-22...

It is possible to include variation in rates of children living in low income families within local authorities over time.

>2/3

of the effectiveness of spending was attributable to decreases in child poverty

Between 2014-15 and 2021-22...

A substantial amount of this was due to the effect of spending on reducing rates of children in poverty: 2.75 per 10,000 of the decrease

However, during this time period, increases in spending tended to be associated with increases in child protection plan rates.

The "cost" of investing in support

'Costs' associated with children being looked after are notoriously difficult to estimate. The moral and human costs of not supporting families should outweigh any economic costs.

A £100 per child increase in spending on preventative services would cost approximately £1.11billion in England.

According to estimates by CPAG, we would expect reductions in child poverty to result in returns of approximately £3.05billion; 275% of the investment.

When spending on preventative services is low, more spending on safeguarding services tended to lead to increased rates of children in care

In 2009-10 and 2010-11, around 28 percent of local authorities had preventative services spending that was £200 or more higher per child than average between 2009-10 and 2021-22

There was also evidence that higher spending on preventative services could attenuate the relationship between increases in child poverty and children in need rates.

I.III Findings: Investing in preventative services was effective for reducing all forms of child welfare interventions before 2013, but only effective for reducing rates of children in care since 2017.

Between around 2013 and 2017, there was no strong evidence that increased spending was associated with decreased rates of intervention.

I.IV Findings: Spending on preventative and supportive services does not have the same impact on rates of intervention everywhere.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Children Looked After rate

In 76.8% of local authorities, we would expect an increase in spending on preventative services to result in a decrease in children looked after rate over two years.

Of these, 60.3% had a probability of direction greater than 0.89, and 19% had a probability of direction between 0.75 and 0.89.

31.1% of local authorities had estimated reductions in CLA rates that were at least twice as large as the median estimated change.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Child Protection Plan rate

In 85.4% of local authorities, we would expect an increase in spending on preventative services to result in a decrease in child protection plan rate over two years.

However, of these, only around one third had a probability of direction greater than 0.89, with around another third having a probability of direction between 0.75 and 0.89.

11.9% of local authorities had estimated reductions in CPP rates that were at least twice as large as the median estimated change.

Local Authority variation in the effect of a £100 per child increase in preventative and supportive services on Children in Need rate

However, of these, 40.2% had a probability of direction greater than 0.89, and 26.8% had a probability of direction between 0.75 and 0.89.

And around 43% of local authorities had estimated reductions in CIN rates that were at least twice as large as the median estimated change.

Children should be able to expect to be supported to the same extent, no matter where in the country they are, to live safely at home.

More effective spending means children's services are more dependent on these services being funded.

Local authorities do not tend to have spending effects that reduce all types of intervention rates to similar extents.

This variation can help inform local policy and practice learning.

Spending effectiveness was not strongly associated with general levels of child poverty.

There was moderate evidence that local authorities that had protected preventative spending had more effective services for reducing children looked after rates, on average.

There was a 93.4% probability that greater reductions in preventative spending between 2009-10 and 2021-22 were associated with less effectiveness per £1 per child invested in prevention.

There was also moderate evidence that local authorities that had higher overall spending had more effective services for reducing children in need rates.

There was a 93.9% probability that higher overall spending between 2009-10 and 2021-22 was associated with greater effectiveness per £1 per child invested in prevention.

I.V Findings: There were substantial inequalities in the effects of preventative spending on child welfare intervention rates across age, gender, and ethnicity.

Between 2009-10 and 2021-22, increases in spending were most strongly associated with decreased rates of intervention among children aged 16 and older, children aged 10 to 15, and children aged under 1.

Indicative changes in rates: -16.9 per 10,000 (16 and older); -4 per 10,000 (10 to 15); -1.7 per 10,000 (Under 1); +0.42 per 10,000 (5 to 9); 1.92 per 10,000 (1 to 4).

Between 2014-15 and 2021-22, the effect for one to four year olds was reversed, and the effect for children aged over 10 was far more muted, potentially due to the removal of the mediating factor of poverty.

The predicted effect of a £100 per child increase in preventative spending varied dramatically between local authorities.

The figures on the right-hand side represent the middle 50 per cent range of the expected effects across local authorities in England.

-11% to +4%
Under 1

-10% to +16%
1 to 4

-10% to +10%
5 to 9

-14% to +2%
10 to 15

-26% to -2%
16 and older

Local authorities tended to have a tendency towards spending that was either effective for reducing rates of care for younger children, or effective for reducing rates of care for older children.

Between 2009-10 and 2021-22, the effect of increased spending on children looked after rates was notably stronger among the male child populations than the female child population.

Though the effect was substantial for both groups of children.

However between 2014-15 and 2021-22, the direct effect of spending on preventative and supportive services in isolation of any mediating effect of poverty was no longer substantial for female children.

But, female children's rates of care within local authorities over time were more strongly associated with increases in child poverty.

There was only slightly more variation across local authorities for male children than there was for female children.

-14% to no change
for male children

-11% to +3%
for female children

0.793

Correlation between effect of spending for female children and effect of spending for male children across local authorities

The effect of spending was more strongly associated with changes in rates of children from minoritised ethnic groups in care.

Between 2014-15 and 2021-22, after removing any mediating effect of poverty, we find that the direct effect of increased spending among most minoritised ethnic groups is no change in rates of care.

Children from minoritised backgrounds faced far more spatial inequality in the effect of spending across local authorities than White children.

3.5 times more variation
for Black children compared to White

3.5 times more variation
for Asian children compared to White

2.3 times more variation
for Mixed Heritage children compared to White

4.3 times more variation
for other ethnicity children compared to White

There was an alarming lack of correlation between the relationship between spending on preventative and supportive services and rates of care across different ethnic populations.

Despite the fact that we might expect the relationship between spending and rates of care to be stronger in urban local authorities, this did not seem to be the case.

I.VI Findings: Investing in Prevention & Support: Summary

There was strong evidence to support the idea that investing in prevention and support effectively reduces rates of child welfare interventions, specifically, rates of children in care or on child protection plans.

Reduced spending on preventative and supportive services can explain a substantial proportion of recent increased in rates of children in care, but increases in child poverty have been more impactful. However, investing in support can also assist with tackling child poverty, and both goals should be pursued in tandem.

Investment in support and prevention is very likely to be at least cost-neutral, and a cursory economic analysis using cost estimates from respected sources suggests investing in these services saves money.

Investing in only safeguarding services is likely to increase rates of children placed in care if services to support families and young people are not also adequately resourced.

However, we must be mindful of both inequalities over time, place, and between different groups. Investing in preventative services may be effective for reducing rates of intervention on average, but such investment needs to be inclusive, and should not depend on when a child is growing up, where they are growing up, or their age, gender, or ethnicity.

This complexity is difficult to manage, and more complexity in the form of intersectional analyses is needed, but mapping out variation across local authorities creates great capacity to learn from one another to address inequalities in the effects of support.

II.I Methods: If spending has such a large effect, why has this not been found in studies before?

The major reason why the findings from this study differ from those of other studies is that a within-between regression model is used, as well as the inclusion of a lagged term.

A groundbreaking study, by Dr. Davara Bennett et al.¹, used these models before to explore differences between one to four year olds and older children.

Standard Regression Model

$\bar{Y} = B_0 + B'X'$

Random Intercepts Multilevel Regression Model

$\bar{Y_{ij}}=B_0+B'X'_{ij}+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Within-between Regression Model

$\bar{Y_{ij}}=B_0+B'_w(X'_{ij}-\bar{X}'_j)+B'_b\bar{X}'_j+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Differences in the substantive interpretation of coefficients:

$B'_w$ — What is the association between changes in $X$ around the mean for groups $j$ and changes in $Y$ around the random intercepts estimated by $U_{0j}$ ?

$B'_b$ — What is the association between group mean differences in $X$ and group mean differences in $Y$ ?

Standard Regression Model

$\bar{Y} = B_0 + B'X'$

Random Intercepts Multilevel Regression Model

$\bar{Y_{ij}}=B_0+B'X'_{ij}+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Within-between Regression Model

$\bar{Y_{ij}}=B_0+B'_w(X'_{ij}-\bar{X}'_j)+B'_b\bar{X}'_j+U_{0j}\\U_{0j}=N(0,\sigma_0)$

Yaremych, et al. 2021. Centering Categorical Predictors in Multilevel Models: Best Practices and Interpretation. Psychological Methods.