Popular in Sociology

Culture and Society

The City as a Growth Machine: Toward a Political Economy of Place

New York University Harvey Moltoch

Culture and Society

Carceral Entrapments: Views from the Prison/Street Interface in India

Indian Institute of Technology Delhi Mahuya Bandyopadhya

Ethnicity and Racism

Were Muslim Barbarians Really Knocking On the Gates of Ontario?

University of Ottawa Natasha Bakht

Culture and Society

Poor People’s Survival Strategies

The University of Texas at Austin Javier Auyero

Ethnicity and Racism

Sociology of Whiteness

Arizona State University Monica McDermott

Ethnicity and Racism

Narratives of Harm and the case for detention abolition

Queen's University Sharry Aiken

Books

The Peer Effect

City University of New York Margaret M. Chin

Pre-problem families: predictive analytics and the future as the present

View more insights in Social Sciences - Sociology - Culture and Society

View more figures from University of Southampton

View the page of Rosalind Edwards

Read the Study

Rosalind Edwards discusses how predictive analytics is seen as a way of identifying the risk of future problems in families.

Image courtesy of interviewee. March 27, 2024

Generating Workspace

Export as PDF

Ross Edwards discusses the implications of using digital administrative data records and predictive analytics in family welfare and child protection services. He explains that predictive analytics involves algorithmic analysis to identify which families may need intervention based on correlations between past behaviors and undesirable outcomes. This video video practice creates a "pre-pro space," where families are anticipated to face potential future problems even if they haven't occurred yet, resulting in a shift from reactive to proactive interventions.

The session highlights parents' views on predictive analytics, noting a lack of attention to their opinions and the ethical concerns this raises. Edward's research, conducted with colleagues, explores parents' assessments of data use in decision-making processes, revealing a mix of acceptance and concerns. While some parents see potential benefits in flagging individuals in distress, they also express strong apprehensions about the accuracy, fairness, and potential biases in the data used, particularly regarding racial disparities.

Parents expressed worry about the accuracy of records, the possibility of inaccurate predictions leading to unjust interventions, and concerns that families could be overlooked or wrongfully flagged. The presentation concludes with recommendations for clearer rights regarding personal data, mechanisms for reporting and correcting errors, and regular reviews of the predictive models' effectiveness. It emphasizes the need for public dialogue on the legitimacy of using administrative data to forecast family issues, advocating for a more ethical and transparent approach.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.325 --> 00:02.385
Hi, my name is Ross Edwards

00:02.385 --> 00:04.865
and I'm gonna explain how the use

00:04.865 --> 00:07.105
of digital administrative data records

00:07.405 --> 00:09.745
and predictive analytics and family welfare

00:10.405 --> 00:12.465
and child protection services, how

00:12.465 --> 00:15.065
that's positioning families in a pre-pro space.

00:15.605 --> 00:18.225
I'm gonna talk about how parents themselves view this,

00:19.045 --> 00:21.745
and I'll be covering what predictive analytics is,

00:22.285 --> 00:25.775
why it's used, what a pre-pro space means.

00:26.195 --> 00:28.615
And I'm also gonna be telling you about parents' views

00:28.795 --> 00:31.495
and I'll say something about the implications for policy.

00:32.875 --> 00:36.615
So what is predictive analytics and why is family data

00:36.835 --> 00:40.375
and predictive analytics and important subject is

00:40.375 --> 00:42.895
because there's been a step change in the practice

00:42.895 --> 00:44.855
of assessing and monitoring families

00:45.555 --> 00:47.535
across many national contexts.

00:47.595 --> 00:50.775
Police and child protection authorities are turning

00:50.835 --> 00:52.215
to algorithmic analysis.

00:52.215 --> 00:55.615
They're doing that in an effort to forecast which

00:56.135 --> 00:58.975
families will need intervention, and that's

00:58.975 --> 01:01.975
because advocates for predictive data modeling argue

01:02.365 --> 01:06.135
that these algorithmic analysis enables streamlined,

01:06.655 --> 01:09.015
efficient, and objective decision making.

01:10.265 --> 01:14.295
Algorithms are a sequence of programmed instructions, rules

01:14.555 --> 01:15.695
and calculations.

01:16.405 --> 01:18.975
They're built so that they can model the tendency

01:19.035 --> 01:21.655
for correlations of particular characteristics

01:22.035 --> 01:25.375
to occur in families who have undesirable outcomes.

01:26.755 --> 01:29.625
Predictive surveillance takes place using algorithms,

01:30.365 --> 01:31.825
and it involves the collection

01:31.925 --> 01:34.945
and the linking together of separate sources

01:35.245 --> 01:38.265
of administrative data that's held on families

01:38.765 --> 01:39.905
and then that joined up.

01:39.905 --> 01:41.145
Data can be used

01:41.145 --> 01:43.865
to identify potential problem behaviors in the future

01:44.255 --> 01:48.355
through algorithmic processing digital sources

01:48.655 --> 01:52.195
of everyday administrative data on families from education,

01:52.655 --> 01:56.675
health benefits, housing, police, et cetera.

01:56.705 --> 01:57.915
They're all joined together

01:58.655 --> 02:01.995
and the modeling algorithm can automatically scan

02:02.375 --> 02:03.595
the merge data sets.

02:04.495 --> 02:07.075
It looks for families that match the patterns of factors

02:07.105 --> 02:09.155
that are indicated for problem families,

02:09.495 --> 02:13.635
and it does that in an effort to identify high, medium,

02:14.175 --> 02:16.795
or low risk of potential family problems.

02:17.615 --> 02:20.595
But it's not just an individual's family data that's used

02:20.595 --> 02:22.515
to forecast their own future actions.

02:23.265 --> 02:26.955
Data from all families is drawn into the predictive modeling

02:27.135 --> 02:29.915
net, net and it's other families behaviors

02:29.915 --> 02:31.155
that determine whether

02:31.175 --> 02:32.755
or not a family is deemed

02:32.775 --> 02:35.115
to pose a futurist to their children.

02:36.625 --> 02:40.085
So algorithms scan the merge data set looking

02:40.185 --> 02:43.845
for correlations that match patterns of factors in families

02:44.175 --> 02:45.725
where undesirable social

02:45.985 --> 02:47.845
and criminological outcomes have

02:47.875 --> 02:49.685
already happened in the past.

02:50.385 --> 02:53.605
And then this patterning is used to identify risk

02:53.605 --> 02:56.485
of harm occurring in other families in the future,

02:56.985 --> 02:58.885
and this creates what's called a pre

02:58.925 --> 02:59.925
Space for families.

03:01.235 --> 03:03.885
This sort of mass digital predictive monitoring

03:03.885 --> 03:04.925
that I've just described

03:04.925 --> 03:06.925
that involves a shift in timeframes.

03:07.855 --> 03:09.315
The established rationales

03:09.335 --> 03:11.755
for state intervention in families on the basis

03:11.895 --> 03:14.195
of what's happening is overturned.

03:14.935 --> 03:18.755
It becomes intervention in families on an anticipatory basis

03:19.295 --> 03:20.875
of what, what hasn't happened,

03:21.175 --> 03:22.755
but might sometime in the future.

03:23.875 --> 03:26.485
This is how families become pulled into a digital

03:27.155 --> 03:31.895
pre-pro space that's taking an imagined hypothetical future

03:32.275 --> 03:34.095
and treats it as if it's definite

03:34.315 --> 03:36.135
and parents are blameworthy

03:36.315 --> 03:38.375
and they're culpable in the present.

03:39.675 --> 03:42.935
The pre-pro space is a temporal paradox.

03:43.715 --> 03:46.375
It suggests simultaneously that something that hasn't

03:46.375 --> 03:48.575
yet occurred is a foregone conclusion.

03:49.465 --> 03:52.325
So that suppose future is bought into the present

03:52.425 --> 03:53.565
and acted upon.

03:54.645 --> 03:57.825
So how might parents feel about information on them

03:58.125 --> 04:00.305
and other families being used predictably?

04:01.425 --> 04:02.995
There's been very little attention

04:03.095 --> 04:05.555
to parents' views on this use of their administrative

04:05.615 --> 04:08.995
and other data services don't have to ask parents consent,

04:09.015 --> 04:10.595
so it's unlikely they know what's happening.

04:11.295 --> 04:14.615
My colleagues Val Giles and Sarah Gorin

04:14.755 --> 04:17.295
and I thought that was a serious ethical

04:17.515 --> 04:18.775
and democratic mission.

04:19.825 --> 04:22.205
So we researched parents' assessments

04:22.225 --> 04:24.685
and concerns about data analytic systems

04:25.025 --> 04:26.405
for service decision making,

04:26.985 --> 04:30.125
and we researched how comfortable they were about families

04:30.175 --> 04:34.125
being positioned in a predictive pre-pro space.

04:35.115 --> 04:38.735
We ran a representative survey and we also held group

04:38.875 --> 04:40.255
and individual discussions

04:40.255 --> 04:43.415
with parents about their views on predictive analytics.

04:44.035 --> 04:45.975
We did that as part of our research project,

04:46.205 --> 04:48.015
exploring parents' understandings

04:48.235 --> 04:51.415
of operational data linkage and use practices.

04:53.715 --> 04:56.695
The parents who took part in our research came from a

04:56.695 --> 04:59.935
diversity of social, racial and occupational backgrounds.

05:00.905 --> 05:03.195
Some parents had experienced family support

05:03.195 --> 05:06.275
and intervention services, but a lot of others haven't.

05:07.535 --> 05:10.395
For the most part, parents from all backgrounds

05:10.455 --> 05:13.755
and experience, they could see merit in algorithm

05:13.915 --> 05:17.445
applications to administrative data that

05:17.555 --> 05:20.405
because it could identify FA individual families

05:20.405 --> 05:21.725
with difficulties in the present

05:22.065 --> 05:23.765
who might mistreat their children.

05:24.145 --> 05:26.365
As one of 'em has said, I think if you're struggling,

05:26.365 --> 05:27.765
it's probably good to be flagged up

05:27.765 --> 05:29.685
because obviously the kids are probably at risk.

05:30.465 --> 05:32.435
But there was a recurrent

05:32.615 --> 05:34.595
and major theme of assertions

05:34.825 --> 05:37.075
that the past is not predictive.

05:37.935 --> 05:40.465
Here parents had major concerns that centered

05:40.465 --> 05:45.025
around accuracy and fairness, stereotyping and prejudice

05:45.765 --> 05:47.705
and determinism and its consequences.

05:49.155 --> 05:52.805
Parents didn't have a lot of confidence in the accuracy

05:52.805 --> 05:55.485
and fairness of the sources of administrative data

05:55.755 --> 05:57.965
that predictive analytics draws upon.

05:58.935 --> 06:02.545
They discussed examples of recorded information

06:02.975 --> 06:06.345
that was based on biased assumptions and judgmental views.

06:06.435 --> 06:09.545
Black parents had especially strong concerns about

06:09.775 --> 06:12.785
unjustified racialized labeling, parents and children,

06:14.485 --> 06:16.545
but either where information is flawed,

06:17.055 --> 06:20.025
data errors often stay on parents' administrative records.

06:21.115 --> 06:24.815
So has the potential for that biased information to be fed

06:25.275 --> 06:29.295
and reed endlessly into algorithmic data analysis

06:29.675 --> 06:31.375
and become part of a vicious cycle

06:31.605 --> 06:35.175
that magnifies in justices in predictive modeling.

06:37.585 --> 06:41.285
The issue of data analytics anticipating problems when in

06:41.285 --> 06:43.445
fact a family not only doesn't have the issues

06:43.745 --> 06:45.765
but wouldn't go on to develop them in the future.

06:46.145 --> 06:49.725
That's known as false positives and false positives.

06:49.725 --> 06:51.925
Worry parents, as one of them said,

06:52.765 --> 06:55.565
I think it's quite dangerous to use someone else's situation

06:55.585 --> 06:58.485
and try and predict someone else because you could be right,

06:58.665 --> 07:00.325
but you could be massively wrong.

07:01.665 --> 07:04.875
Parents were also bothered about families being passed over

07:05.015 --> 07:06.275
by predictive analytics

07:06.275 --> 07:08.395
because they didn't fit the data profile

07:08.775 --> 07:10.115
of dysfunctional families

07:10.185 --> 07:12.115
that the algorithm was working with.

07:13.065 --> 07:14.845
That's known as false negatives.

07:15.465 --> 07:18.435
Parents often mentioned affluent middle class families.

07:18.435 --> 07:22.665
In this respect, the idea of positioning parents

07:22.765 --> 07:24.825
and children in a pre-pro space

07:25.445 --> 07:27.465
and pulling together the past of the future

07:28.095 --> 07:30.585
into the present in a deterministic way

07:30.895 --> 07:32.665
that made parents quite indignant.

07:33.135 --> 07:35.865
They talked about the potential damage that could be caused

07:35.925 --> 07:37.825
by the surveillance and investigation

07:38.125 --> 07:40.185
and maybe children being removed from parents

07:40.245 --> 07:41.585
and taken interstate care.

07:42.765 --> 07:46.295
Some parents mentioned dystopian scenarios like terminated

07:46.355 --> 07:48.255
two stuff and big brother watching.

07:48.475 --> 07:50.695
They'd said that to convey their discomfort.

07:52.585 --> 07:56.405
But parents could also be skeptical about how families

07:56.585 --> 07:59.325
who are pulled into pre-pro space, were going

07:59.325 --> 08:02.365
to receive interventions in the face of austerity

08:02.745 --> 08:05.005
and diminish funding for service provision

08:05.265 --> 08:06.925
and staff shortages.

08:08.355 --> 08:10.055
In the UK and many other places,

08:10.145 --> 08:12.655
there have been severe cuts to public spending.

08:13.755 --> 08:15.325
It's one of the parents pointed out

08:16.275 --> 08:18.775
the reality is services have been cut till.

08:18.775 --> 08:21.255
There's nothing there to say that you are going to put

08:21.255 --> 08:24.575
that information into an algorithm to identify needs.

08:24.575 --> 08:26.975
When there are no services for them to access,

08:27.085 --> 08:28.535
there's no help they can have

08:29.005 --> 08:31.055
that feels really disingenuous.

08:33.085 --> 08:35.225
So what ought to happen about authorities use

08:35.445 --> 08:36.465
of predictive modeling

08:36.845 --> 08:38.625
and the pre-pro space, it creates,

08:39.925 --> 08:44.185
at the very least parents' apprehensions about the atri

08:44.205 --> 08:47.425
and prejudice data that can be fed into predictive models

08:48.315 --> 08:50.375
and their worries about the determinism

08:50.515 --> 08:52.455
and accuracy of predictive analytics

08:52.555 --> 08:54.575
and the very least that should be taken seriously.

08:55.785 --> 08:57.575
We've called for three main provisions.

08:58.785 --> 09:00.285
One is that there should be clear

09:00.505 --> 09:02.685
and enforceable rights for individuals

09:03.105 --> 09:06.725
to view the personal data that's held about them online.

09:07.225 --> 09:08.965
In administrative databases,

09:10.295 --> 09:14.195
two parents should have a proper system that allows them

09:14.395 --> 09:16.955
to report errors in the information held about them

09:17.465 --> 09:21.075
with speedy investigations and correction of data errors.

09:22.105 --> 09:25.435
Finally, there needs to be a regular independent review

09:25.695 --> 09:28.035
and publication of the accuracy

09:28.295 --> 09:30.555
and the utility of the predictive models

09:30.985 --> 09:33.315
that are being applied to administrative data.

09:35.065 --> 09:38.385
Ultimately though, we need a serious public conversation

09:38.445 --> 09:42.305
to take place about the legitimacy of state construction

09:42.605 --> 09:46.105
of a pre-pro space using administrative data

09:46.285 --> 09:47.585
that's held on families.

00:00 - Hi, my name is Ross Edwards and I'm gonna explain how the use of digital administrative data records and predictive analytics and family welfare and child protection services, how that's positioning families in a pre-pro space.

00:15 - I'm gonna talk about how parents themselves view this, and I'll be covering what predictive analytics is, why it's used, what a pre-pro space means. And I'm also gonna be telling you about parents' views and I'll say something about the implications for policy. So what is predictive analytics and why is family data and predictive analytics and important subject is because there's been a step change in the practice of assessing and monitoring families across many national contexts.

00:47 - Police and child protection authorities are turning to algorithmic analysis. They're doing that in an effort to forecast which families will need intervention, and that's because advocates for predictive data modeling argue that these algorithmic analysis enables streamlined, efficient, and objective decision making. Algorithms are a sequence of programmed instructions, rules and calculations. They're built so that they can model the tendency

01:19 - for correlations of particular characteristics to occur in families who have undesirable outcomes. Predictive surveillance takes place using algorithms, and it involves the collection and the linking together of separate sources of administrative data that's held on families and then that joined up. Data can be used to identify potential problem behaviors in the future through algorithmic processing digital sources

01:48 - of everyday administrative data on families from education, health benefits, housing, police, et cetera. They're all joined together and the modeling algorithm can automatically scan the merge data sets. It looks for families that match the patterns of factors that are indicated for problem families, and it does that in an effort to identify high, medium, or low risk of potential family problems. But it's not just an individual's family data that's used

02:20 - to forecast their own future actions. Data from all families is drawn into the predictive modeling net, net and it's other families behaviors that determine whether or not a family is deemed to pose a futurist to their children. So algorithms scan the merge data set looking for correlations that match patterns of factors in families where undesirable social and criminological outcomes have

02:47 - already happened in the past. And then this patterning is used to identify risk of harm occurring in other families in the future, and this creates what's called a pre Space for families. This sort of mass digital predictive monitoring that I've just described that involves a shift in timeframes. The established rationales for state intervention in families on the basis

03:11 - of what's happening is overturned. It becomes intervention in families on an anticipatory basis of what, what hasn't happened, but might sometime in the future. This is how families become pulled into a digital pre-pro space that's taking an imagined hypothetical future and treats it as if it's definite and parents are blameworthy and they're culpable in the present. The pre-pro space is a temporal paradox.

03:43 - It suggests simultaneously that something that hasn't yet occurred is a foregone conclusion. So that suppose future is bought into the present and acted upon. So how might parents feel about information on them and other families being used predictably? There's been very little attention to parents' views on this use of their administrative and other data services don't have to ask parents consent, so it's unlikely they know what's happening.

04:11 - My colleagues Val Giles and Sarah Gorin and I thought that was a serious ethical and democratic mission. So we researched parents' assessments and concerns about data analytic systems for service decision making, and we researched how comfortable they were about families being positioned in a predictive pre-pro space. We ran a representative survey and we also held group and individual discussions

04:40 - with parents about their views on predictive analytics. We did that as part of our research project, exploring parents' understandings of operational data linkage and use practices. The parents who took part in our research came from a diversity of social, racial and occupational backgrounds. Some parents had experienced family support and intervention services, but a lot of others haven't. For the most part, parents from all backgrounds and experience, they could see merit in algorithm

05:13 - applications to administrative data that because it could identify FA individual families with difficulties in the present who might mistreat their children. As one of 'em has said, I think if you're struggling, it's probably good to be flagged up because obviously the kids are probably at risk. But there was a recurrent and major theme of assertions that the past is not predictive.

05:37 - Here parents had major concerns that centered around accuracy and fairness, stereotyping and prejudice and determinism and its consequences. Parents didn't have a lot of confidence in the accuracy and fairness of the sources of administrative data that predictive analytics draws upon. They discussed examples of recorded information that was based on biased assumptions and judgmental views. Black parents had especially strong concerns about unjustified racialized labeling, parents and children,

06:14 - but either where information is flawed, data errors often stay on parents' administrative records. So has the potential for that biased information to be fed and reed endlessly into algorithmic data analysis and become part of a vicious cycle that magnifies in justices in predictive modeling. The issue of data analytics anticipating problems when in fact a family not only doesn't have the issues but wouldn't go on to develop them in the future. That's known as false positives and false positives.

06:49 - Worry parents, as one of them said, I think it's quite dangerous to use someone else's situation and try and predict someone else because you could be right, but you could be massively wrong. Parents were also bothered about families being passed over by predictive analytics because they didn't fit the data profile of dysfunctional families that the algorithm was working with. That's known as false negatives.

07:15 - Parents often mentioned affluent middle class families. In this respect, the idea of positioning parents and children in a pre-pro space and pulling together the past of the future into the present in a deterministic way that made parents quite indignant. They talked about the potential damage that could be caused by the surveillance and investigation and maybe children being removed from parents and taken interstate care.

07:42 - Some parents mentioned dystopian scenarios like terminated two stuff and big brother watching. They'd said that to convey their discomfort. But parents could also be skeptical about how families who are pulled into pre-pro space, were going to receive interventions in the face of austerity and diminish funding for service provision and staff shortages. In the UK and many other places, there have been severe cuts to public spending.

08:13 - It's one of the parents pointed out the reality is services have been cut till. There's nothing there to say that you are going to put that information into an algorithm to identify needs. When there are no services for them to access, there's no help they can have that feels really disingenuous. So what ought to happen about authorities use of predictive modeling and the pre-pro space, it creates,

08:39 - at the very least parents' apprehensions about the atri and prejudice data that can be fed into predictive models and their worries about the determinism and accuracy of predictive analytics and the very least that should be taken seriously. We've called for three main provisions. One is that there should be clear and enforceable rights for individuals to view the personal data that's held about them online. In administrative databases,

09:10 - two parents should have a proper system that allows them to report errors in the information held about them with speedy investigations and correction of data errors. Finally, there needs to be a regular independent review and publication of the accuracy and the utility of the predictive models that are being applied to administrative data. Ultimately though, we need a serious public conversation to take place about the legitimacy of state construction of a pre-pro space using administrative data

09:46 - that's held on families.

Faculti AI can make mistakes