Modelling migration probabilities - School of Geography

Sample of Anonymised Records: User Meeting
Propensity to migrate by ethnic group:
1991 & 2001
Paul Norman 1, John Stillwell 2 & Serena Hussain 2
School of Geography, University of Leeds
1
ESRC RES-163-25-0032
2
ESRC RES-163-25-0028
Internal migration of Britain's ethnic populations
John Stillwell & Serena Hussain
• Understand how migration intensities and patterns vary
between ethnic groups at various geographical levels in 2001
and what changes can be identified between 1991 & 2001
What happens when international migrants
settle? Ethnic group population trends &
projections for UK local areas under alternative
scenarios
Phil Rees, Paul Norman & Peter Boden
• Investigate trends in migration propensities by ethnic group
& need age-specific migration rates for projections
Research questions …
• Do propensities to migrate vary by ethnic group, age, sex
at the national level and for sub-national areas?
• How do spatial patterns of internal ethnic group migration
vary with spatial distribution of ethnic populations and
immigration at LAD scale, 2000-01?
• What are the characteristics of the origins and
destinations of ethnic group migrants in 2000-01 at ward
level?
• What other factors influence migration propensities?
• Have there been any changes between 1991 & 2001?
Specification …
Stillwell & Hussain have commissioned tables from ONS,
one of which is for flows at LAD scale in England & Wales:
Age
White
Indian
Pakistani &
other South
Asian
Chinese
African, Caribbean,
Black British &
other Black
Mixed
0-15
16-19
20-24
25-29
30-44
45-59
60+
Ethnic groups consistent with ethnic group data in SMS
Reasonable geographical detail but lacks information on:
• sex & other factors & is only for 2000-01
Other
Specification …
Data sources: 1991 & 2001 Individual Licensed SAR
Study population: household & communal establishment
residents in England & Wales, excluding recent immigrants
Close as possible to the ethnic group and age detail from the
commissioned table, plus sex, migrant status and region
(GOR)
Derive variables consistent for 1991 & 2001 that the literature
shows to be related to the propensity to migrate:
• Social Class
• Educational achievement (degree)
• Tenure (owners, public renters, private renters, communal)
• Employment status (active, unemployed, other)
• Health (reported limiting long-term illness)
Deriving variables … Ethnicity
Aim
1991 Indiv. SAR
2001 Indiv. SAR
Best fit
• White
• White
• British
• Irish
• Other White
• White
• Indian
• Indian
• Indian
• Indian
• Pakistani & other
South Asian
• Pakistani
• Bangladeshi
• Pakistani
• Bangladeshi
• Pakistani & other
South Asian
• Chinese
• Chinese
• Chinese
• Chinese
• African, Caribbean,
Black British & other
Black
• Black African
• Black Caribbean
• Black Other
• Black Caribbean
• Black African
• Other Black
• African, Caribbean,
Black British & other
Black
• Not available
• White & Black Caribbean
• White & Black African
• White & Asian
• Other Mixed
• Other
• Mixed
• Other
• Other Asian
• Other-Other
• Other Asian
• Other
Deriving variables … Social Class
Available in 1991, emulated in 2001 using NS-SEC
• Age (0-15; 16-19; 20-24; 25-29; 30-44; 45-59; 60+)
• Migrant status (year before census) (y/n)
• Educational achievement (degree)
• Tenure (owners, public renters, private renters)
• Employment status (active, unemployed, other)
• Health (reported limiting long-term illness)
Deriving variables … Government Office Region
Available in 2001 SAR, needs estimating for 1991
1991
‘areap’
Link to GOR via
Centroid
2001
‘GOR’
GOR added to 1991 individual records using ‘areap’ link
Contextual information
1991: economic depression, high unemployment
2001: economically buoyant, low unemployment
Regional migration rates, different pattern
1991 migration rates
2001 migration rates
North East
7.79
9.10
North West
7.80
8.63
Yorkshire & The Humber
8.52
9.37
East Midlands
8.50
8.67
West Midlands
7.56
7.82
East of England
9.27
7.78
London
10.45
9.69
South East
10.06
8.27
South West
10.25
9.48
Wales
7.65
8.43
England & Wales
8.97
8.71
Government Office Region
Modelling migration: logistic regression
Outcome variable
• Dichotomous categorical outcome: Did / Did not migrate
• Model predictions to lie between 0 & 1
Explanatory variables
• Categorical or continuous
Model outputs
• Odds of event compared with a ‘base’ or ‘reference’ level
• Can be expressed as probabilities
Dale A, Fieldhouse E, Holdsworth C (2000) Analyzing Census Microdata. Arnold: London
Series of models developed
• Simple then adding in various factors
• ‘Interactions’ explored
Modelling migration probabilities … (model n)
Influences on migration (year before census)
continued …
Modelling migration probabilities … (model n)
Influences on migration (year before census)
continued …
Modelling migration probabilities … (model n)
Influences on migration (year before census)
continued …
Modelling migration probabilities … (model n)
Modelling summary …
Ethnicity
• South Asian groups less likely to migrate than White group
• White group propensities higher in 2001 than 1991
• Chinese, Black & Other groups higher odds of migration in
1991, but less difference from White group by 2001
Other factors consistent with literature
Government Office Region: complex picture
Rates in NE
from 7.79%
to 9.10%
Rates
1991
Rates
2001
Using the model outputs …
Modelled migration probabilities by ethnic group
For forecasting scenario models
• Estimate single year of age & trends
• Corroborate from other sources
SARs goods & less than goods …
SARs, < goods …
• Inconsistencies between 1991 & 2001 extracts, e.g. variable
definitions, geography & different decisions by UK’s NSAs
• Delays for 2001 release, loss of momentum
• For 2011 need continuity from previous censuses
• Loss of variable detail down to lowest in common?
SARs microdata a great resource
• Large sample size
• Extraction of study population of interest
• Derivation of variables & versatile crosstabulations
• Comparisons between 1991 & 2001 & then 2011
• Modelling techniques