Determinants of Mobile Phone Penetration: Panel Threshold Evidence from Sub-Saharan Africa

ABSTRACT

Despite the evolving literature on the development benefits of mobile phones, we still know very little about factors that influence their adoption. Using 25 policy variables, we investigate determinants of mobile phone penetration in 49 Sub-Saharan African countries with data for the period 2000–2012. The empirical evidence is based on contemporary and non-contemporary OLS, Fixed Effects, System GMM, and Quantile Regression techniques. The determinants are classified into six policy categories. They are: (i) macroeconomic, (ii) business/bank, (iii) market-related, (iv) knowledge economy, (v) external flows, and (vi) human development. Results are presented in terms of threshold and non-threshold effects. The former has three main implications. First, there are increasing positive benefits in regulation quality, human development, foreign investment, education, urban population density, and Internet penetration. Second, there is evidence of decreasing positive effects from patent applications. Third, increasing damaging influences are established for foreign aid and return on equity. Non-threshold tendencies are discussed. Policy implications are also covered with emphasis on policy syndromes to enhance more targeted implications for worst-performing nations.

KEYWORDS:

• Panel data
• Mobile phones
• Development
• Africa

JEL CLASSIFICATION:

• C23
• L96
• O11
• O33
• O55

Notes

^1. Mobile, mobile phones, mobile telephony, and cell phones are used interchangeably throughout this study.

^2. The positioning of the study also steers clear of recent African business literature on the use of information technology for doing business (Afutu-Kotey, Gough, & Owusu, 2017; Kuada, 2009, 2014, 2015; Tchamyou, 2017), knowledge for the successful implementation of projects (Hashim, 2014; Ika & Saint-Macary, 2014; Joseph, Erasmus, & Marnewick, 2014; Ofori, 2014), and reducing information asymmetry that is related to business transactions (Tchamyou & Asongu, 2017).

^3. For example, the following sets of variables do not enter into the same specifications: IRS & LDS, DSav & HCE, SSE & Internet, SSE & Credit, SSE & HDI, Patent & Credit, SSE & Bbrchs, IRS & Inflation, SSE & Popg, Internet & Bbrchs, HDI & Bbrchs and HDI & Internet. “Interest Rate Spread,” “Net Interest Margin,” and “Lending Deposit Rate” cannot all enter into the same specification because of concerns about mulitcollinearity. Only two of the variables can be employed in a given specification. In Table 5, “Interest Rate Spread” is not used because “Net Interest Margin” and “Lending Deposit Rate” are used. In Table 6, “Interest Rate Spread” is not used because “Net Interest Margin” and “Lending Deposit Rate” are used. In Table 7, “Interest Rate Spread” is used with either “Net Interest Margin” or “Lending Deposit Rate.” In the light of these clarifications, the need to avoid concerns about multicollinearity justifies the multitude of specifications in the empirical results section.

^4. “First, the null hypothesis of the second-order Arellano and Bond autocorrelation test (AR[2]) in difference for the absence of autocorrelation in the residuals should not be rejected. Second the Sargan and Hansen overidentification restrictions (OIR) tests should not be significant because their null hypotheses are the positions that instruments are valid or not correlated with the error terms. In essence, while the Sargan OIR test is not robust but not weakened by instruments, the Hansen OIR is robust but weakened by instruments. In order to restrict identification or limit the proliferation of instruments, we have ensured that instruments are lower than the number of cross-sections in most specifications. Third, the Difference in Hansen Test (DHT) for exogeneity of instruments is also employed to assess the validity of results from the Hansen OIR test. Fourth, a Fischer test for the joint validity of estimated coefficients is also provided” (Asongu & De Moor, 2017, p. 200).