Page 72 - IJPS-9-2

International Journal of
Population Studies Relationship between population aging and innovativeness

Norway, the Philippines, Poland, Portugal, Romania, We also performed an alternative estimation by
South Africa, Spain, Sweden, Switzerland, Thailand, implementing a two-step GMM system estimation with
Turkey, United Kingdom, and United States. While the standard errors corrected to account for the small size of
choice of countries may have been somewhat constrained our sample. It should be noted that the two-step estimator
by data availability, the dataset was considered sufficiently is more efficient than the one-step GMM system. However,
heterogeneous in terms of institutional settings and standard errors tend to be downward biased in small
demographic conditions. The heterogeneity in data was samples. To consider this, Windmeijer (2005) proposed
crucial for examining the relationship between population a finite sample correction to estimate the variance in this
aging and innovation across different contexts. By including linear dynamic model. Therefore, we applied Windmejer’
countries with diverse institutional frameworks and s correction in the two-step system GMM estimator. We
demographic characteristics, the analysis could capture a also estimated a model including the number of trademark
wide range of factors that may influence the innovativeness applications per 1,000 inhabitants in the right-hand side
of nations. This heterogeneity strengthens the robustness of Equation I (labeled as one-step GMM’). This variable is
of the analysis and enables more nuanced insights into the included as a proxy of the level of creativity in the country
relationship under investigation. (Williams & McGuire, 2010; Flikkema et al., 2019). Given

2.2. Methods that its inclusion implies a reduction of the sample size
due to missing observations for some years/countries, we
To answer RQ1 and RQ2, we estimated the following decided not to include it in our baseline model. All the
dynamic data panel model: statistical analyses are carried out using STATA 17.

Pat =ρPat +βx +v +ε (I)
it i,t−1 it i it 3. Results
Where Figure 2 shows the evolution of patent applications in
( |ε x i ,1985 1989 ,… E , x i ,2015 2019 , ) = v i 0 (II) 100,000 residents from 1985 to 2019 in our sample. It
it
Pat is the number of patent applications per 1,000 is essential to clarify that the scale used in Figure 2A–D
it
inhabitants of country i at time t (t = 1985 – 1989,…, 2015 differs from the dependent variable used in Equation
– 2019). Given that innovation is generally an incremental I as it is meant for visual purposes. All analyses were
process, we allowed a certain degree of persistence by indeed conducted using the number of patents per 1,000
including a lag of the dependent variable in our empirical residents. To allow readability, we divided the figure into
model. Rho is the coefficient associated to the lagged four panels (low, low-medium, medium-high, and high
dependent variable. The explicative variables x are the share innovative countries) based on the following criteria: 2019
it
of population over the age of 65 (as our proxy of country patent applications for inhabitants (PAI) in country i ≤ the
aging), the life expectancy at birth, the natural growth 2019 cross-country first quartile; the 2019 cross-country
rate of the population, the net migration rate, the share of median < 2019 PAI in country i ≤ the 2019 cross-country
population aged 25–64 who have had tertiary education, median; the 2019 cross-country median < 2019 PAI in
the security and property rights protection index, the country i ≤ the 2019 cross-country third quartile; 2019 PAI
flexibility in the business regulation index, and a dummy in country i > the 2019 cross-country third quartile.
equal to one when the country is classified as a high-income It is interesting to note that one of the most innovative
country by the World Bank. Beta represents the vector of and, at the same time, the oldest country in the world,
coefficients associated to our explicative variables. The v i such as Japan, is experiencing a declining trend in the
are the panel-level effects. By construction, we consider number of patent applications in the last 20 years. Only
that the lag of the dependent variable is endogenous the Republic of Korea exhibits an increasing trend in
given that it will be correlated with v , making the most the number of applications in the whole period under
i
common estimators (for instance, ordinary least squares) analysis.
inconsistent. The model can be consistently estimated
through the Arellano-Bover/Blundell-Bond Generalized Figure 3 shows the percentages of the population over
Method of Moments (GMM) system estimator (Arellano & the age of 65 in 2019 for each country under consideration.
Bover, 1995; Blundell & Bond, 1998), which is designed to Note that all the countries that are currently in the highest
deal with panels with few periods and larger cross-section quartile in the number of patents application are also,
units (our necessity to group data in 5-year intervals led except for China and South Korea, characterized by a
us to have only seven periods). The method assumes that larger share of population over 65 and, at the same time,
no autocorrelation exists in the idiosyncratic errors ε (this are experiencing a flattening or a decline in the number
it
can be tested through the Arellano-Bond test). of submissions to the patent office. To explain this decline,

Volume 9 Issue 2 (2023) 66 https://doi.org/10.36922/ijps.0429