Page 90 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.83, # 1, 2026, pp. 82-106

We do not‚ for example‚ control for GDP growth or educational level in the main
regression as no sufficient data were available. Alternatively‚ one could control for
the business cycle or structural breaks. For example‚ higher GDP growth could
decrease unemployment and‚ at the same time‚ increase remittances (if exchange rates
change or the emigrant diaspora's incomes increase). These problems are potential
sources of endogeneity. There are several business cycles in the period we cover. The
long panel with fixed effects reduces the bias. However‚ because we do not control
for everything‚ we must be clear that our regression results are associations consistent
with theory‚ rather than conclusive causal evidence.

SPSS Procedure
The panel data is entered in SPSS in long format (with the variables country‚ year‚
remittance%‚ and unemployment% for each year-country pair). As shown in Table 1‚
Frequencies/Descriptives functions were used to calculate the means and standard
deviations of each variable. The Bivariate Correlation dialog was used to obtain
correlation statistics‚ and the Linear Regression dialog was used to obtain regression
statistics. Using the filtered datasets for each individual country‚ we estimated the
regression of unemployment on remittances. For the pooled model‚ we created dummy
variables for each country‚ and included these dummies plus remittances as independent
variables. We checked at each stage of our analysis‚ for example using residual plots‚
for the residuals to behave normally: none of these were considerably heterogeneous‚
which is partly because we aggregated the data by year for this model. Because our data
are annual and there are a limited number of observations per country‚ we rely on
substantive effect sizes and consistency with our theoretical expectations.

Following the reporting of the results of the analyses in tables‚ we will report the
correlation coefficients‚ regression coefficients and importance levels‚ below the
relevant tables which will make the visualisation and interpretation of the results
easier. We will not overly round the results‚ reporting two or three decimal places
when appropriate. Statistical importance was evaluated using standard conventions‚
noting whether p-values were below the 0.05 threshold for statistical importance.

Each of these three methods of analysis (descriptive‚ correlation and regression) could
be used to determine if remittances offset or compensate the labor market effects of
emigration in those countries. They would compensate the labor market if the
countries that receive most remittances have low or decreasing unemployment. If
these regressions were successful‚ the relationship would be statistically negative.
Otherwise‚ if remittances are high‚ and even possibly increasing‚ unemployment
might be high (or increasing)‚ resulting in a zero or positive correlation of remittances
to unemployment. This leads to the multi-country approach to see how countries could

90