The Challenge

• Consistent with National and Regional Tier 1 GDP
• Detailed data available for re-use
• Source code and method open for critique and improvement
• Easy interactive access

The data

• Business Demography Statistics
• Linked Employer - Employee Database
• Regional GDP
• National GDP

The method (bottom up)

• reconcile the various earnings measures
• scale up employee numbers to earnings
• scale up earnings to GDP

The method (alternative, top down version)

• reconcile the two GDP measures
• allocate any ambiguity on basis of earnings
• allocate any remaining ambiguity on basis of employees

Commuter correction

• earnings only published by place of residence
• use the Census 2013 to reallocate earnings on basis of commuting
• known area for future improvement

Assumptions

• Earnings a good indicator of value added
• Employee numbers a good indicator of earnings
• No interaction between TA and earnings:GDP ratio for specified industry
• No interation between Region/TA and inter-industry relative earnings:GDP ratios of industries
• Commuter patterns unchanging and not industry-specific
• Price movements not industry-specific
• Price movements not region-specific

Question

Do either of this variables in 2005 have an impact on the next ten years of growth:

• Agricultural importance
• GDP per capita

Modelling approach

• Generalized additive model
• Degrees of freedom allocated to variables based on strength of Spearman \(\rho^2\)
• Smoothed two dimensional spline to control for spatial correlation

Strength of relationships

\[ GDPGrowth_i = \beta_0 + \beta_1 \times PropAg2005 +\\ \beta_2 \times GDPpp2005 + \beta_3 \times s(lat, long, 5) + \epsilon_i\]

• no evidence of an agriculture in 2005 effect
• no evidence of a GDP per capita in 2005 effect
• strong spatial effect

The spatial effect