

<?xml version="1.0" encoding="UTF-8"?>
<record>
  <title>Global Dynamics of Online-Learning Search Interest: An Advanced Multivariate Time-Series Analysis Across 20 Countries (2004-2021)</title>
  <journal>Journal of E-Technology</journal>
  <author>Darya-Anna Kaybiyaynen</author>
  <volume>17</volume>
  <issue>4</issue>
  <year>2026</year>
  <doi>https://doi.org/10.6025/jet/2026/17/4/163-175</doi>
  <url>https://www.dline.info/jet/fulltext/v17n4/jetv17n4_1.pdf</url>
  <abstract>This study examines the temporal evolution and cross-country interdependence of public search interest in
â€œOnline Learningâ€ using a balanced monthly panel covering 20 countries from January 2004 to October
2021. The dataset contains 214 observations per country and 4,280 country month observations in total,
with no missing values. The analysis combines univariate and multivariate time series procedures to
characterize stationarity, autocorrelation, dynamic dependence, long run relationships, and the propagation
of shocks. Augmented Dickey Fuller (ADF) and Kwiatkowski Phillips Schmidt Shin (KPSS) tests were applied
to all series in levels and first differences. Eight series satisfied the combined stationarity criterion in levels,
whereas 18 series satisfied it after first differencing. Autocorrelation and partial autocorrelation functions
were examined up to 12 lags, followed by ARIMA and monthly seasonal SARIMA modeling for focal nonstationary
series. For multivariate analysis, a parsimonious six country system comprising Argentina, India,
Indonesia, Iran, Mexico, and the United States was selected to avoid excessive parameterization. Informationcriterion
selection supported a VAR(3) specification, while the Johansen procedure indicated a cointegration
rank of five, motivating estimation of a VECM. Granger causality testing with false discovery rate adjustment
identified 14 significant directional relationships. Impulse response functions show that most shocks are
strongest at short horizons and generally attenuate over time, while forecast error variance decomposition
reveals heterogeneous dependence on own and cross country innovations. Overall, the findings demonstrate
that online learning search interest is both country specific and dynamically interconnected, with meaningful
short run transmission and substantial long run linkage among the focal series.</abstract>
</record>
