Need of care in interpreting Google Trends-based COVID-19 infodemiological study results: potential risk of false-positivity

Table 2 The number of significant keywords which have temporal association with the COVID-19 positivity trend

	Maximum lag in VAR model	Granger causality		Pearson correlation		Spearman’ rank correlation
	Maximum lag in VAR model	(A) raw p-value < 0.05	(B) FDR < 0.05	(C) raw p-value < 0.05	(D) FDR < 0.05	(E) raw p-value < 0.05	(F) FDR < 0.05
AU	4	21	16	24	20	35	34
	6	21	16	24	20	35	34
	8	21	16	24	20	35	34
CA	4	25	20	29	25	37	37
	6	28	24	29	25	37	37
	8	28	24	29	25	37	37
GB	4	22	18	39	38	42	42
	6	16	8	39	38	42	42
	8	20	10	39	38	42	42
IE	4	17	10	23	18	26	25
	6	19	15	23	18	26	25
	8	20	15	23	18	26	25
IN	4	4	1	49	49	53	53
	6	4	1	49	49	53	53
	8	5	1	49	49	53	53
JP	4	10	5	33	33	44	44
	6	12	5	33	33	44	44
	8	14	5	33	33	44	44
SG	4	10	5	21	18	24	24
	6	11	7	21	18	24	24
	8	11	6	21	18	24	24
US	4	29	26	38	34	41	40
	6	27	27	38	34	41	40
	8	28	25	38	34	41	40
ZA	4	7	3	33	31	37	36
	6	11	3	33	31	37	36
	8	12	4	33	31	37	36

The number of keywords in which GT trend had a significant Granger-caused COVID-19 positivity trends (A, raw p-value < 0.05; B, FDR < 0.05), the number of keywords whose GT trend had significant Pearson correlation with the COVID-19 positivity trends (C, raw p-value < 0.05; D, FDR < 0.05), and the number of keywords whose GT trend had significant Spearman’s rank correlation with the COVID-19 positivity trends (E, raw p-value < 0.05; F, FDR < 0.05). For Granger causality in each region, the lag order of the VAR model is varied in the range of 1–8
In addition, inversely, the number of keywords of which trend is predictable by the COVID-19 positivity trend was very few (not shown in this table): p-value / FDR are shown in the ‘Granger (opposite)’ columns in Additional files 3 and 4, where FDR < 0.05 was only 1 in Great Britain, 2 in India, 1 in Japan, and at most 11 in South Africa
Abbreviations: AU Australia, CA Canada, GB Great Britain, IE Ireland, IN India, JP Japan, SG Singapore, US United States, ZA South Africa, FDR False discovery rate, VAR Vector autoregression

ISSN: 1471-2288