1 Applying Circular Applying Circular Statistics to the Statistics to the Study of Graduate Job Study of Graduate Job Search: The Case of Search: The Case of Great Britain. Great Britain. Alessandra Faggian Alessandra Faggian 1 , Jonathan Corcoran , Jonathan Corcoran 2 and and Philip McCann Philip McCann 3 1 School of Geography, University of Southampton, UK School of Geography, University of Southampton, UK 2 UQ Social Research Centre University of Queensland, Australia UQ Social Research Centre University of Queensland, Australia 3 Management School, The University of Waikato, NZ Management School, The University of Waikato, NZ 48th ERSA Congress 48th ERSA Congress , 27 , 27 th th – 31 – 31 st st August 2008, Liverpool, UK August 2008, Liverpool, UK
22
Embed
1 Applying Circular Statistics to the Study of Graduate Job Search: The Case of Great Britain. Alessandra Faggian 1, Jonathan Corcoran 2 and Philip McCann.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Applying Circular Statistics to Applying Circular Statistics to the Study of Graduate Job the Study of Graduate Job Search: The Case of Great Search: The Case of Great Britain.Britain.
Alessandra FaggianAlessandra Faggian11, Jonathan Corcoran, Jonathan Corcoran22 and Philip McCann and Philip McCann33
11School of Geography, University of Southampton, UKSchool of Geography, University of Southampton, UK22UQ Social Research Centre University of Queensland, AustraliaUQ Social Research Centre University of Queensland, Australia
33Management School, The University of Waikato, NZManagement School, The University of Waikato, NZ
48th ERSA Congress48th ERSA Congress, 27, 27thth – 31 – 31stst August 2008, Liverpool, UK August 2008, Liverpool, UK
Introduction
Role of space on the graduate labour market How does human capital investment affect
graduate migration patterns? What are the geographical characteristics of
the UK graduate job search areas? How do University, personal and regional
characteristics affect these movements?
Theoretical framework Job search (e.g. Lippman and McCall 1976a,b and 1979) and
human capital (Sjaastad, 1962) theories both predict that the radius of the job search area increases with an increase in human capital. So, assuming jobs are randomly distributed over space:
Lower HK
Higher HK
Radius Expansion (all directions)
Theoretical framework
BUT, are jobs are randomly distributed over space? Aren’t higher quality/higher wage jobs more likely to be concentrated in urban areas rather than randomly scattered in space?
If this is the case, shouldn’t we observe a different pattern for people with different human capital levels?
In the case of University graduates, shouldn’t we observe different patterns between ‘high achievers’ (1 or 2.1) and ‘low achievers’ (2.2 or below)
Theoretical framework
Lower HK
Higher HK
• We cannot talk about ‘radius’ of search.
• The assumption of jobs being randomly distributed over space is a better approximation for less qualified job seekers
Theoretical framework Traditionally, the summary measure used to describe
graduate movements is the ‘average distance’ (linear measure) moved after graduation (which is a proxy for the search radiussearch radius). So, according to expectations:University Average distance
moved by graduates (meters)
Percentage of "migrants"
Ranking(2001)
The University of Cambridge
149,841 96.9 1
The University of Oxford
136,978 96.5 2
The University of Southampton
121,832 92.8 11
Strathclyde University 64,522 59.9 51
Liverpool Hope University
74,216 56.7 88
Thames Valley University
52,027 53.7 123
BUT, how do we capture the ‘shape’ of the job search area?
We need some non-linear (circular) (circular) measures, which allow use to identify: The ‘direction’‘direction’ of graduate movements on top
of the average distance moved using data at University level (circular average) (circular average)
The ‘spread’‘spread’ of graduate movements (circular (circular variance)variance)
7
Theoretical framework
A “curious byway of statistics…somewhere between the analysis of linear and the analysis of spherical data” (Fisher 1993, p.1)
Deal with directional data (either a compass direction or some unit of time)
Early roots date back to the mid-eighteenth century (Bernoulli, 1734)
Why circular measures???
Methodology: Circular measures
Consider a hypothetical origin zone with 5 moves each with a single student centred around a northerly direction; 3400, 3500, 80, 100, 230
´Origin zone
34003500 80 100
230
Individual journey to a job destination
(i) Circular measures
Computing the standard linear mean (340 + 350 + 8 + 10 + 23 / 5) equates to 146.20, or around a south-easterly direction – the fallacy of the linear measure!
´Origin zone
34003500 80 100
230
Individual journey to a job destination
Linear mean
(i) Circular measures
11
Circular mean and circular varianceCircular mean and circular variance
n
iiA
1
)sin(
n
iiB
1
)cos(
180/
and
where
2
1221
1 BAn
R
1
1
arctan( / ) 0
arctan( / ) ) 0
A B if B
A B if B
And the circular variance R-bar (spread) is:
Applying the circular mean to the hypothetical example given above, equates to 2.230 or just around north. Circular variance = 0.035 (i.e. low spread)
´Origin zone
34003500 80 100
230
Individual journey to a job destination
Linear mean
Circular mean
(i) Circular measures
Compute for each origin zone and classify into sectors for thematic mapping Number of sectors
determined by data 4 sectors 8 sectors etc
337.50 to 22.50
8 sectors (450 slices)
(i) Circular measures
Data
~12 million observations on students in the academic years between 1995/96 and 2005/06 (Source: Students Data by HESA) HESA)
~1.5 million observations on graduates jobs (Sources: First Survey Destination, 95-00 and Destination of Leavers in Higher Education by HESA, 2002-05)
Focus in this paper: 1999/2000 cohort (~300,000 observations)
15
1. Average direction of graduates by HEI
250 250
250
250
200 200
200
200
150 150
150
150
100 100
100
100
50 50
50
50
0
90
180
270
16
2. ‘Spreads’: examples…UNIDIRECTIONALITY
250 250
250
250
200 200
200
200
150 150
150
150
100 100
100
100
50 50
50
50
0
90
180
270 250 250
250
250
200 200
200
200
150 150
150
150
100 100
100
100
50 50
50
50
0
90
180
270
Mean Vector (µ) 214.975°Variance (R) 0.357
Average direction
Mean Vector (µ) 214.975°Variance (R) 0.357
Mean Vector (µ) 215.307°Variance (R) 0.347
15km TTWA 50km
HEI n.114:low variance
Different definitions Different definitions of non-migrantof non-migrant
50 50
50
50
40 40
40
40
30 30
30
30
20 20
20
20
10 10
10
10
0
90
180
270
17
MULTIDIRECTIONALITY
95% confidence
intervalAverage direction
15km TTWA 50km
50 50
50
50
40 40
40
40
30 30
30
30
20 20
20
20
10 10
10
10
0
90
180
270 40 40
40
40
30 30
30
30
20 20
20
20
10 10
10
10
0
90
180
270
Mean Vector (µ) 54.218°Variance (R) 0.743
Mean Vector (µ) 67.111°Variance (R) 0.744
Mean Vector (µ) 35.652°Variance (R) 0.543
HEI n.90: high variance
Different definitions of ‘non-migrant’Results for the 15km radius area and TTWAs are incredibly similar, with the only exception of HEIs located along the border between two TTWAs (in which case the 15Km might even be preferable)
3. Determinants of spread of movements
Based on previous work on graduate migration in GB (Faggian and McCann 2006, Faggian et. al 2006, 2007a and b)
and on human capital migration theory and gravity type models, we expect the circular variancevariance to be related:
• Selectivity of HEI attended• Degree classification obtained• Age of graduates• Degree of specialisation of HEI attended• Spatial constraints of HEI location (coast)• London attraction• Previous migratory behaviour (distance moved from home to HEI)