Page 27 - IJPS-7-2

International Journal of
Population Studies Gender gap in life expectancy in South and East Europe

acquired by inversion of the consistent estimator of the E( ) (8)
,
variance-covariance matrix of the moment conditions. As tj tj it, it j,
further stated by Wooldridge (2001a), if there are m > k + 1
total moment conditions, where, k represents the number of Where, m is the test statistics of j-th order of serial
j
covariates in the model, the weighting matrix would have m correlation and ρ is the average j-th order auto-covariance.
j
× m dimensions. In addition, if concerns are existed about 3. Results
heteroskedasticity, there are methods for calculating standard
errors and testing statistics that are robust to heteroskedasticity 3.1. Trends of the gender gap in LEAB in South and
of not known form. The usual approach when facing East European countries
heteroskedasticity of unknown form is to use the GMM Solid evidence has shown that a spectacular rise in LEAB
(Baum et al., 2003). The GMM estimator using weighting for South and East European countries that started
matrix places no restrictions on either the unconditional since 1965 has continued to progress and some of South
or conditional (on Z) variance matrix of u, that is, the European countries (e.g., Spain, Italy, Malta, Portugal, and
i
i
asymptotically efficient estimator can be obtained without Cyprus) are now the leaders of LEAB in Europe (Caselli
making additional assumptions (Wooldridge, 2001b). First, et al., 2014; Eurostat, 2021; United Nations, 2022a).
differencing is applied to remove the unobserved effect, and
thereafter, lags are utilized as instrumental variables for the Figure 1A and 1B presents LEAB levels and the gender
differenced lagged dependent variable. Wooldridge (2001a) gaps in LEAB for 24 countries in the study region. The figures
added further that since the original time-varying errors show that the group of South and East European countries
are supposed to be serially uncorrelated, the differenced is heterogeneous in different levels of LEAB. In general,
errors must have serial correlation. GMM is well fitted for LEAB for both sexes is lower for Russia, Ukraine, Belarus,
securing efficient estimators that consider for the serial Moldova, Türkiye, Bulgaria, Romania, Montenegro,
correlation (e.g., Arellano and Bond (1991). As specified by Bosnia and Herzegovina, Macedonia, Serbia, and
Lucas et al. (1997), the approach of Arellano and Bond (1991) Hungary, whereas it is higher in Italy, Portugal, Spain,
supports consistent estimates under very weak distributional Malta, Cyprus, Greece, and Slovenia. The group of
assumptions. Developed by Arellano and Bond (1991), countries as Albania, Croatia, Czechia, Slovakia, and
GMM techniques can control for both unobserved country- Poland is more or less in an intermediate position. In
specific effects and first difference non-stationary variables, South Europe, not only the gains in LEAB in the period
and further to overcome the problem with endogeneity of 1991 – 2020 vary greatly but geographical inequalities are
the explanatory variables using instruments and test for the more pronounced than elsewhere. During the period of
presence of autocorrelation (Adusei, 2013). For dynamic our research study (1991 – 2020), Türkiye, Croatia, and
model estimation using yearly data, it is found that GMM with Malta are the three countries where the expected LEAB
additional moment conditions can provide more accurate increased the most (by 9.5, 8.1, and 7.8 years, respectively),
estimates than two-stage least squares (Wooldridge, 2001a). unlike Montenegro and Bulgaria, where the increase was
only 0.5 years and 3.7 years, respectively.
For models estimated by GMM, the first- and second-
order serial correlation statistics initiated by Arellano One characteristics of mortality in the study period
and Bond (1991) were used as a method to test for serial 1991 – 2020 are the stagnation of LEAB in 1991 – 1995
correlation. The test has in effect two different statistics, due to the increase in probability of dying at almost all
one for first-order correlation and the other for second- age groups that were resulted from impacts of collapse
order correlation. If the innovations are i.i.d., it is expected of previous socioeconomic systems in these countries.
the first-order statistic to be significant (with a negative During the beginning of the 1990s, many East European
autocorrelation coefficient), and the second-order statistic countries even underwent decrease in LEAB, while other
to be not significant (Arellano & Bond, 1991). The test parts of Europe witnessed an increase (SORS, 1997). The
statistics proposed by Arellano and Bond is calculated as considerable increase of LEAB in the mid- or late-1990s
shown in Eqs. (6-8): is probably the response of that earlier stagnation in
mortality trends that are generally attributed to an end of

m j (6) health crises of the transitional period. As Avdeev et al.
j
VAR () (2011) stated, across Europe, females have a longer life
j
than males, but the range of the gap in LEAB between
the two sexes varies extensively, as does the development
1 T

T 3 j t 4 j tj (7) in the region throughout the time. Thus, South Europe
j
follows the same trend of East Europe but with greater
Volume 7 Issue 2 (2021) 21 https://doi.org/10.36922/ijps.v7i2.389