Factors affecting the magnitude and timing of temporary moves in Australia: A Poisson Regression Analysis
Presentation to the 2007 International Population Geographies Conference, Hong Kong, 10-13th July 2007
Elin Charles-EdwardsDominic BrownMartin Bell
• Study background – Service population estimates (Cook 1996; Lee
1999)• Those persons who demand goods or services from
providers… (s)uch persons may be permanent or temporary residents of an area (Cook 1996)
Background
Service populations
Permanent residents (ERP)
Daytime populationTemporary Residents
Temporary mobility: those moves more that one night in duration that do not entail a change in usual
residence.
Lower bound: 24 hrsUpper bound: 12 months
• Estimating temporary populations
1. Direct• Census –temporal
resolution• Travel surveys (NVS)-
spatial resolution• Expensive and time
consuming
2. Indirect• Accommodation
surveys–visitors in private dwellings
• Symptomatic data (e.g. electricity, water usage) – accessibility, benchmarking
• (e.g. Smith 1989, Happel et. al 2002)
3. Simulation• Based on the underlying dimensions of temporary
population mobility
Background
Duration
Periodici
ty
Circuits Distance
Connectivity
Impact
FrequencySeasonality
Magnitude
Tem
pora
l
Spatial
Number of visitors?When do they arrive?How long do they stay?
Background
Seasonality: the systematic intra-year variation in visitation caused by exogenous factors (e.g. climatic), institutional factors (e.g. timing of public holidays) or a combination of the two.
What do we know?
• Few large scale studies of temporary population mobility
• No accepted conceptual framework within which to situate this mobility
• Currently no scientific theory of visitor seasonality
•Tourism literature has identified a number of different causes of tourism seasonality
• Natural (e.g. Climate)• Institutional (e.g. School Holidays)• Calendar effects (e.g. Easter)
•How do we start thinking about temporary mobility and the ways in which it varies through space and time?
What will get us there?
Origin Destination
DistanceWeather
Climate
Populationsize
Economic function
School Holidays
Weather
Climate
Populationsize
Economic function
School Holidays
Harvest Calendar
Festivals
Tim
e
1. Scale – spatial and temporal2. Fully saturated model – sparsely populated
Business cycles
DiasporaDiaspora
Data •National Visitor Survey
• Comprehensive source of data of temporary population mobility in Australia
• Continuous sample ~80 000 persons per annum• Variables: destination, origin, timing, purpose and
duration of visit/trip• Sampling variability
• Precludes the direct estimation of temporary visitors to small regions
• Precludes use of fully saturated model
•Dependent variable - monthly inflows to 68 Australian Tourism Regions
What will get us there?
What will get us there?
2005 ASGC Tourism Regions
±
0 500 1,000 1,500250Kilometres
I nsuffi cient counts
Data: Explanatory Variables
What will get us there?
Determinants Time Origin Destination
Events Day X
Public holidays Day X
School holidays Week/Month
X
Temperature Month X X
Precipitation Month X X
Sunlight hours Month X X
Harvest calendars Month X X
Business Cycles Month X
Population size Annual X X
Economic function Annual X X
Accessibility Annual X X
Diaspora Annual X X
Model 1 Model 2
Model type Time series regression Cross sectional regression
Question What causes the number of visitors to a particular destination to vary over time (daily, weekly, monthly)?
What factors underlie the spatial distribution of temporary moves at time t?
Assumption Assumes factors affecting the magnitude of temporary moves are invariant over time
Assumes factors affecting the magnitude of temporary moves are invariant across space
Regression type Poisson Poisson
Number of models
68 12
Geography Individual Tourism Regions All Tourism Regions
What will get us there? - Models
-Approach separates model into temporal and spatial components
• Run stepwise Poisson Regression Models • Model 1 (68 time series models)
• Dependent variable – monthly inflows to Tourism Region
• Independent variables – max. temp, sunshine hrs, precipitation
• Offset- monthly inflows to all Tourism Regions• Apportionment model
• Model 2 (12 cross-sectional models)• Dependent variable- inflows to all tourism regions• Independent variables – max. temp, sunshine hrs,
precipitation, Tourism Quotient, ARIA score• Offset- Estimated Resident Populations• In-migration rate model
What will get us there?Methodology
Results – Model 1
• Model fits are poor overall• Independent variables in 26/68 models accounted
> 50 per cent of null deviance (G2 > 50 per cent)• 12 models had a G2 statistic greater than 70 per
cent
Selected Results – Model 1
Model 1Region G^2 Variable Deviance
% Null deviance Coefficient
Relative Risk
Eyre Peninsula 86.6 Intercept -4.869096
(Coastal) Max Temperature 11733 82.8 0.008658 0.9
Precipitation 154 1.1 -0.017464 -1.7
Sun Hours 380 2.7 -0.057488 -5.6
Phillip Island 77.5 Intercept -5.62E+00
(Coastal) Max Temperature 42532 75.6 3.73E-02 3.8
Precipitation 160 0.3 -2.05E-01 -18.5
Sun Hours 900 1.6 4.96E-02 5.1
Outback Qld 74.2 Intercept -4.09E+00
(Inland) Max Temperature 39862 56.0 -5.79E-02 -5.6
Precipitation 11684 16.4 -3.68E-02 -3.6
Sun Hours 1271 1.8 1.43E-01 15.4
Snowy Mountains 70.7 Intercept -5.60E+00
Max Temperature 110564 59.7 -9.72E-02 -99.6
Precipitation 18900 10.2 1.38E-02 1.4
Sun Hours 1480 0.8 1.63E-01 17.6
• Poor model fits overall suggest that key determinants are missing from the model
• Temperature accounts for most of the deviance in these models – direction of effect varies
• Snowy Mountains (-ve)• Phillip Island (+ve)
• Precipitation accounts for a moderate proportion of deviance for a number of regions- direction of effect varies
• Snowy Mountains (+ve)• Outback QLD (-ve)
Results – Model 1
Results – Model 2
• Model fits are good overall
50.0
55.0
60.0
65.0
70.0
75.0
80.0
85.0
90.0
95.0
100.0
Janu
ary
Feb
ruar
y
Mar
ch
Apr
il
May
June
July
Aug
ust
Sep
tem
ber
Oct
ober
Nov
embe
r
Dec
embe
r
Month
G2
stat
isti
c
Results – Model 2
Month G^2 Variable Deviance% Null deviance Coefficient RR
January 85.1 Intercept -1.52
ARIA Score 544623 11.9 0.12 12.6
Max. Temperature 742986 16.3 -0.02 -2.0
Tourism Quotient 2494038 54.7 1.92 578.7
October 67.3 Intercept -2.22
ARIA Score 539836 21.9 0.16 17.2
Max. Temperature 116947 4.8 -0.01 -1.3
Tourism Quotient 995149 40.4 1.21 234.3
• Tourism Quotient accounts for most of the model deviance for all months (+ve) followed by ARIA score (+ve)
• Maximum monthly temperature is the only factor varying at a monthly scale accounting for even moderate amounts of deviance
Conclusions
• Early stages of research – major findings• Models work for some types of regions more
than others• Not a common set of factors that apply to all
regions• Easier to model baseline flows• Time series model better captures the seasonal
variation in flows
• Where to next?– Need to refine conceptual framework – Include more explanatory variables – difficult!!!– Disaggregate by purpose of trip?– Reintegration of temporal and spatial dimensions?