Colonial Origins of Comparative Development in Ghana

Abstract

A striking feature of Ghana’s development landscape is the stark development disparity between a relatively developed South and a trailing North. Explanations for the disparity have often been hinged on differences in geography and past colonial experience. In this study, I provide an empirical justification for the historical hypothesis that the dynamics of colonial rule contributed significantly to the development divergence between the North and the South. I exploit the asymmetric regional distribution of past colonial public investments in education, health and infrastructure to show that the dynamics of colonial rule explain a significant portion of the development disparity between the two regions. I also survey compelling historical anecdotes to show that prior to the colonial project the North was a relatively prosperous region.

Keywords:

• Colonial rule
• development disparity
• colonial investments
• Northern Ghana
• Southern Ghana

Notes

1 Using survey data, I show that there is a strong visual correlation between light density at night and access to public goods at the district level. The survey data also shows that Northern households are relatively poorer.

2 The terms light density at night, luminosity, and nighttime light mean the same and are used interchangeably.

3 To the best of my knowledge, 2013 is the most recent year for which the light density data is available from the DMSP-OLS.

4 In the specifications where I add cell-level controls, the number of observations reduces to 2085. The missing 6 are “sea cells” so their data could not be easily extracted using the tools of ArcGIS. Since they are all southern cells, hence relatively more developed, the development gap would likely be slightly underestimated.

5 This approach is adopted from Michalopoulos and Papaioannou (2013). who employ it because of the presence of zeros in the luminosity data. Zero in the luminosity data denotes cases where either the place has no electricity, or the place is so dim that the satellite sensors could not capture any light. Nighttime light is also blank for natural landmarks like lakes.

6 The map is sourced from Map and Geospatial Spatial Information Centre, Lewis Library, Princeton University: https://earthworks.stanford.edu/catalog/princeton-b8515r88g. Date last accessed: 12/08/2021.

7 Though the dummies for sea and border cells are controls for economic geography, they also control for differences in sizes of cells since cells that border the sea and/or another country may measure smaller than 11 km x 11 km.

8 Regions here should not be confused with Northern Ghana and Southern Ghana. Regions as used here are the first subnational administrative units after the central government. There are currently 16 regions in Ghana, following new demarcations in 2019 which created additional 6 regions from the former 10 regions. I use the former 10 regions in this study.

9 Since the outcome variable is log-transformed, percentage change in the dependent variable when the predictor variables record a one unit increase or change from 0 to 1 is more accurately calculated as [exp(B) – 1]*100 (Wooldridge, 2018, p.227).

10 Unless otherwise indicated, all regression coefficients are reported in absolute values.

11 Afrobarometer is a not-for-profit, independent body that conducts public opinion survey in Africa.

12 Assisted schools were schools that depended on the colonial government for their financial maintenance.

13 https://geonode.wfp.org/layers/geonode%3Awld_trs_railways_wfp#more Date last accessed: 06/04/2020

14 The data on the number of assisted colonial schools in 1919 is from (Thomas, 1974).

15 An alternative interpretation is that colonial investments lasted longer period. For example, a school in 1901 is likely to be functional in 1931.

16 This identification framework is heavily drawn from Jedwab and Moradi (2016).

17 Standardizing accounts for changes in mean and standard deviations of population over time.

18 Note that z-score of population at time t is regressed on all opened railway lines as at time t. Thus z-score of 1931 population, for example, is regressed on all lines opened as at 1931, and z-score of 1948 population regressed on all lines opened as at 1948, and so on. All lines were constructed and opened in the colonial era.

19 These are population census years in both colonial and postcolonial eras. For 1960, 1948 and 1901, only urban population data are available.

20 I do not control for 1891 population because the North lacks this data.

21 Absolute average log nighttime light is 3.265.

22 Contemporary factors are postcolonial population variables, distance to the capital city and access to modern road network. Colonial-era factors are the colonial-era population variables.

23 Distance to colonial class one road is not statistically significant. The author did not include the result tables for the robustness checks to save space. They can be provided upon request.

24 The North lacked colonial railways and class one roads, so their effects cannot be estimated in the North sample.

25 The points discussed here are heavily taken from (Huillery, 2009).

26 Donaldson (2018) uses a similar approach to study the effects of colonial railway lines in India. This approach assumes that the two set of cells have similar observables and economic potentials. In terms of economic profitability of the northern lines, feasibility studies had reported that the lines were economically viable (Tsey, 1986) and the revenue could cover the cost of construction within a year of operation (Brukum, 1998).

27 The absolute log mean nighttime light is 2.63 for this sample.

28 I use absolute distance (km), rather than distance dummies. This choice is purely based on convenience as it eases the task of stacking the results into one table.