Uncovering the Time-Frequency Relationship Between Google Trends and COVID-19 Vaccination Metrics: A Hybrid ARDL-Wavelet Coherence Model for Prediction

TL;DR

This paper uses Google Trends and a new statistical model to better predict vaccination rates during the pandemic, showing how online search behavior can help track public health trends.

Contribution

A novel hybrid ARDL-wavelet coherence model is introduced to improve vaccination prediction accuracy and analyze dynamic relationships with Google Trends.

Findings

Google Trends shows significant long-term correlation with U.S. vaccination rates.

The ARDL-wavelet model achieves lower prediction errors (RMSE and MAE) than previous studies.

Wavelet analysis reveals that Google Trends' predictive power weakened as vaccination rates stabilized.

Abstract

Background/Objectives: With the continuous evolution of global infectious diseases such as the coronavirus disease 2019 (COVID-19) pandemic, vaccination remains one of the most effective means of intervention. However, in the process of vaccination promotion, public vaccination behavior often lags behind policy deployment. Accurately predicting vaccination trends early has become one of the key problems in the current public health field. At present, the public's online search (such as Google Trends (Google LLC, Mountain View, CA)) has become an important predictor of vaccination intervention. However, existing studies have low prediction accuracy and a lack of understanding of the dynamic heterogeneity between Google Trends and vaccination behavior. This study aims to improve the prediction accuracy of vaccination using the autoregressive distributed lag model (ARDL model) and solve the dynamic heterogeneity problem using wavelet coherence analysis.

Methods: This paper first uses the ARDL model to study the long-term cointegration relationship between the search index (Google Trends) and the COVID-19 vaccination rate in the United States (including the number of vaccination doses per week, the number of people who receive at least one dose of vaccine per week, and the number of people who complete the full vaccination per week). Then, in order to reveal the time-varying characteristics of the predictor variables, the wavelet coherence method is introduced to study the dynamic interaction between Google Trends and vaccination behavior in the time-frequency domain. There are only a few such relevant studies in the existing literature on vaccination prediction. The data for this study covers December 2020 to May 2022. Vaccination data comes from the US CDC, and Google Trends data comes from keywords extracted from COVID-19 vaccines.

Results: The results of the time domain study show that Google Trends has a significant long-term cointegration relationship with the weekly vaccination rate. Because this paper uses an ARDL model with automatically selected optimal lag orders, the prediction accuracy of the search index for vaccination has been significantly improved. After comparing the prediction accuracy with existing similar studies, the root mean square error (RMSE) and mean absolute error (MAE) obtained in this paper are much smaller than those reported in existing studies. The results of the frequency domain wavelet coherence study show that in the early stage of rapid vaccine promotion (especially in early 2021), Google Trends has a strong correlation with vaccination behavior. From the end of 2021 to 2022, the coherence between GT and vaccination behavior gradually weakened.

Conclusions: The results show that Google Trends has good long-term prediction ability in the vaccine promotion stage, but its short-term prediction effect shows volatility with time and frequency. The ARDL-wavelet coherence analysis framework proposed in this paper provides a new research paradigm for evaluating the public's influence on vaccination. The research results emphasize the potential and importance of Google Trends in the construction of real-time public health response and early warning systems.

Figures