A longitudinal analysis of adherence to COVID-19 preventive measures, media use, and risk perceptions in a Dutch population-based sample

Figures & data

Figure 1. Conceptual diagram of the hypothesized random intercept cross-lagged panel model (RI-CLPM) for the relationships between media use, risk perception, and adherence to the COVID-19 preventive measures. Random intercepts (RIs) are specified by the repeated measures of the latent factors as its indicators with all factor loadings fixed to 1 (Mulder & Hamaker, 2021). Residuals of the within-person variables within Waves 2–4 are allowed to correlate, but paths are not displayed for the sake of visual clarity. w1 = Wave 1 (May). w2 = Wave 2 (June). w3 = Wave 3 (July). w4 = Wave 4 (October).

Table 1. Factor validity statistics to test for longitudinal measurement invariance of the latent variables.

	Media use			Risk perception			Adherence
	Configural	Metric	Scalar	Configural	Metric	Scalar	Configural	Metric	Scalar
Fit statistics
χ²	5753.79	5824.70	6063.67	1673.74	1683.90	2286.96	557.86	589.98	845.60
df	1598	1631	1673	205	217	238	134	146	158
p Value	<.001	<.001	<.001	<.001	<.001	<.001	<.001	<.001	<.001
CFI	0.900	0.899	0.895	0.916	0.916	0.883	0.965	0.963	0.943
TLI	0.889	0.891	0.888	0.887	0.894	0.864	0.950	0.952	0.931
RMSEA	0.047	0.047	0.047	0.078	0.076	0.086	0.052	0.051	0.061
SRMR	0.101	0.101	0.101	0.116	0.116	0.207	0.056	0.059	0.061
Relative fit
Δχ²	–	70.91	238.97	–	10.16	603.06	–	32.125	255.620
Δdf	–	33	42	–	12	21	–	12	12
Δp-value	–	.050	<.001	–	.710	<.001	–	.009	<.001
ΔCFI	–	.001	.004	–	.000	.033	–	.002	.020
ΔTLI	–	.002	.003	–	.007	.030	–	.002	.021
ΔRMSEA	–	.000	.000	–	.002	.010	–	.001	.010
ΔSRMR	–	.000	.000	–	.000	.091	–	.003	.002

Notes: χ²: Chi-square; df: degrees of freedom; p value: chi-square probability; CFI: comparative fit index; TLI: Tucker-Lewis index; RMSEA: root mean square error of approximation; SRMR: standardized root mean squared residual. Relative fit indices (denotes with Δ) indicate changes in these fit statistics when comparing more constrained models to less constrained models (i.e., the metric model was compared to the configural model, and the scalar model was compared to the metric model).

Table 2. Means (SDs) and zero-order correlations for media use, risk perception, and adherence.

	Wave 1	Wave 2	Wave 3	Wave 4
Media use	2.22 (1.16)	2.04 (1.13)	1.92 (1.16)	2.03 (1.15)
Traditional media	2.93 (1.55)	2.70 (1.54)	2.52 (1.56)	2.74 (1.56)
Online news	2.43 (2.07)	2.22 (1.95)	2.11 (1.90)	2.18 (1.96)
Health-related sources	0.67 (1.18)	0.63 (1.09)	0.60 (1.11)	0.61 (1.12)
Social media	2.01 (2.44)	1.79 (2.34)	1.71 (2.30)	1.70 (2.25)
Risk perception	3.44 (0.52)	3.34 (0.52)	3.35 (0.51)	3.49 (0.52)
Perceived susceptibility	2.98 (0.65)	2.86 (0.65)	2.88 (0.64)	3.09 (0.65)
Perceived severity	4.37 (0.71)	4.31 (0.70)	4.30 (0.69)	4.30 (0.74)
Adherence	4.24 (0.68)	4.15 (0.72)	4.07 (0.76)	4.20 (0.70)
Wash your hands often	4.38 (0.83)	4.30 (0.87)	4.21 (0.90)	4.23 (0.90)
Cough and sneeze into your elbow	3.93 (1.26)	3.89 (1.25)	3.87 (1.25)	3.95 (1.21)
Use paper tissues to blow your nose and discard them after use	3.78 (1.47)	3.77 (1.47)	3.71 (1.47)	3.86 (1.42)
Do not shake hands	4.79 (0.69)	4.66 (0.89)	4.61 (0.90)	4.77 (0.70)
Keep 1.5 m distance from others	4.31 (0.80)	4.12 (0.88)	3.93 (0.94)	4.18 (0.83)
Work from home as much as possible	3.64 (1.71)	3.41 (1.74)	–	–
Zero-order correlations
Adherence with risk perception	.22***	.22***	.22***	.24***
Adherence with media use	.17***	.17***	.16***	.20***
Risk perception with media use	.11***	.09**	.11***	.13***

Notes: Risk perception could range from 1 (strongly disagree) to 5 (strongly agree) and adherence could range from 1 (never) to 5 (always), with higher scores indicating higher levels of risk perception and adherence. Media use could range from 0 (never) to 7 (7 days a week), where higher scores indicate more frequent use of that media source. Wave 1 = May; Wave 2 = June, Wave 3 = July; Wave 4 = October. No descriptives available for Wave 3–4 for the last adherence item, as this measure was lifted as from July 2020 (Wave 3). Zero-order correlations were calculated based on the total mean scales of media use, risk perception, and adherence.

Table 3. Standardized estimates of the random intercepts cross-lagged panel models (RI-CLPMs) regarding the between-person and within-person correlations between media use, risk perception, and adherence.

	Traditional media				Online news				Health-related sources				Social media
	β	SE	p	95% CI	β	SE	p	95% CI	β	SE	p	95% CI	β	SE	p	95% CI
Between-person correlations
Media use and risk perception	.06	.02	.075	−.00, .07	.17	.02	< .001	.07, .15	.06	.01	.106	−.00, .04	.01	.03	.702	−.04, .06
Media use and adherence	.25	.03	< .001	.14, .24	.14	.03	< .001	.07, .19	.16	.02	< .001	.04, .10	.05	.04	.210	−.03, .12
Risk perception and adherence	.35	.01	< .001	.06, .10	.35	.01	< .001	.06, .10	.35	.01	< .001	.06, .10	.35	.01	< .001	.06, .10
Within-person correlations
Media use and risk perception
Correlation at w1	.05	.02	.407	−.02, .05	.11	.02	.017	.01, .09	.06	.01	.191	−.01, .05	.02	.02	.685	−.04, .06
Residual correlation at w2	.02	.02	.731	−.02, .03	−.02	.02	.653	−.05, .03	−.01	.01	.866	−.02, .02	.02	.02	.671	−.04, .06
Residual correlation at w3	.11	.02	.091	−.01, .06	.03	.02	.597	−.03, .05	.10	.01	.028	.00, .05	.06	.02	.290	−.02, .07
Residual correlation at w4	.08	.02	.152	−.01, .05	.05	.02	.255	−.02, .06	.04	.01	.429	−.02, .04	.07	.02	.126	−.01, .08
Media use and adherence
Correlation at w1	.01	.02	.879	−.04, .05	.03	.03	.624	−.04, .06	−.05	.02	.371	−.05, .02	.11	.03	.027	.01, .12
Residual correlation at w2	.04	.02	.582	−.03, .06	−.02	.03	.715	−.07, .05	.03	.02	.605	−.03, .05	−.04	.04	.525	−.09, .05
Residual correlation at w3	.04	.02	.331	−.02, .05	.03	.02	.602	−.04, .06	.05	.01	.245	−.01, .04	−.05	.03	.269	−.09, .03
Residual correlation at w4	.05	.02	.295	−.02, .05	.10	.02	.025	.01, .09	.04	.02	.425	−.02, .04	.04	.03	.381	−.03, .08
Risk perception and adherence
Correlation at w1	.03	.01	.711	−.01, .02	.03	.01	.669	−.01, .02	.03	.01	.688	−.01, .02	.03	.01	.646	−.01, .02
Residual correlation at w2	−.02	.01	.666	−.02, .01	−.02	.01	.681	−.02, .01	−.03	.01	.640	−.02, .01	−.02	.01	.757	−.02, .01
Residual correlation at w3	.01	.01	.872	−.01, .02	.01	.01	.919	−.01, .02	.01	.01	.905	−.01, .02	.01	.01	.858	−.02, .01
Residual correlation at w4	.13	.01	.016	.00, .04	.13	.01	.019	.00, .04	.13	.01	.016	.00, .04	.13	.01	.017	.00, .04

Notes: All results but between-person correlation reflect correlations at the within-person level. Results are considered significant at p < .0125 to correct for potential alpha inflation due to multiple testing. w1 = Wave 1 (May). w2 = Wave 2 (June). w3 = Wave 3 (July). w4 = Wave 4 (October).

Table 4. Standardized estimates of the random intercepts cross-lagged panel models (RI-CLPMs) regarding the autoregressive and cross-lagged within-person paths between media use, risk perception, and adherence.

	Traditional media				Online news				Health-related sources				Social media
	β	SE	p	95% CI	β	SE	p	95% CI	β	SE	p	95% CI	β	SE	p	95% CI
Autoregressive paths
Media use
w1 → w2	.06	.06	.316	−.06, .17	.13	.05	.016	.02, .20	.14	.07	.075	−.01, .26	.14	.05	.010	.03, .23
w2 → w3	.14	.07	.015	.03, .29	−.00	.06	.979	−.12, .12	.07	.08	.332	−.08, .23	.10	.06	.115	−.02, .21
w3 → w4	.04	.06	.523	−.08, .15	−.07	.07	.330	−.22, .07	−.17	.08	.030	−.32, −.02	−.09	.08	.263	−.23, .06
Risk perception
w1 → w2	.16	.05	.004	.05, .25	.15	.05	.004	.05, .25	.15	.05	.005	.04, .25	.15	.05	.005	.05, .25
w2 → w3	.08	.06	.223	−.05, .20	.08	.06	.214	−.04, .20	.07	.06	.277	−.06, .19	.07	.06	.257	−.05, .19
w3 → w4	.01	.07	.903	−.12, .14	−.01	.06	.934	−.13, .12	.00	.07	.969	−.13, .13	.00	.06	.959	−.12, .13
Adherence
w1 → w2	−.02	.08	.747	−.19, .14	−.02	.09	.778	−.19, .15	−.02	.09	.774	−.20, .15	−.03	.09	.718	−.21, .14
w2 → w3	.17	.06	.002	.06, .30	.18	.06	.002	.07, .30	.18	.06	.002	.07, .30	.18	.06	.002	.07, .31
w3 → w4	.22	.05	< .001	.11, .31	.22	.05	< .001	.11, .31	.22	.05	< .001	.11, .31	.22	.05	< .001	.11, .31
Cross-lagged paths
Media use → risk perception
w1 → w2	.01	.02	.916	−.04, .04	.06	.01	.182	−.01, .04	.04	.02	.353	−.02, .05	.03	.01	.512	−.01, .02
w2 → w3	.07	.03	.355	−.03, .09	−.02	.02	.788	−.04, .03	.06	.02	.265	−.02, .06	.04	.01	.478	−.01, .03
w3 → w4	−.09	.03	.188	−.08, .02	−.05	.02	.398	−.05, .02	−.09	.03	.182	−.09, .02	−.09	.01	.088	−.05, .00
Media use → adherence
w1 → w2	.09	.03	.111	−.01, .10	.07	.02	.158	−.01, .05	.06	.03	.271	−.02, .08	.02	.01	.628	−.02, .03
w2 → w3	.05	.03	.335	−.03, .09	.05	.02	.410	−.02, .06	.05	.02	.224	−.02, .07	−.01	.02	.925	−.03, .03
w3 → w4	.01	.02	.918	−.04, .05	.06	.02	.179	−.01, .06	.03	.02	.518	−.03, .06	−.03	.01	.517	−.04, .02
Risk perception → adherence
w1 → w2	−.02	.07	.701	−.15, .10	−.02	.07	.759	−.15, .11	−.02	.07	.707	−.16, .11	−.01	.07	.903	−.14, .12
w2 → w3	.01	.07	.825	−.12, .15	.01	.07	.877	−.12, .14	.01	.07	.815	−.12, .15	.01	.07	.799	−.12, .15
w3 → w4	.08	.06	.081	−.01, .24	.08	.06	.093	−.02, .23	.07	.06	.105	−.02, .23	.08	.06	.082	−.01, .24
Risk perception → media use
w1 → w2	−.00	.12	.947	−.24, .22	−.09	.15	.057	−.57, .01	−.04	.10	.381	−.28, .11	.01	.19	.784	−.32, .42
w2 → w3	−.02	.12	.674	−.28, .18	−.03	.15	.495	−.40, .19	.01	.10	.831	−.17, .21	.01	.19	.784	−.22, .43
w3 → w4	.09	.19	.242	−.15, .59	−.03	.24	.667	−.58, .37	.05	.15	.376	−.16, .43	.05	.25	.347	−.26, .74
Adherence → media use
w1 → w2	−.01	.13	.934	−.26, .24	−.05	.18	.397	−.50, .20	−.04	.13	.442	−.34, .15	.02	.23	.724	−.37, .53
w2 → w3	.07	.09	.130	−.04, .32	.07	.13	.189	−.08, .41	.09	.09	.037	.01, .36	−.05	.15	.257	−.45, .12
w3 → w4	.01	.10	.798	−.16, .21	.09	.14	.119	−.06, .48	.02	.08	.615	−.11, .19	−.04	.17	.448	−.45, .20
Adherence → risk perception
w1 → w2	−.10	.05	.096	−.20, .02	−.10	.05	.091	−.20, .01	−.10	.05	.109	−.19, .02	−.10	.06	.095	−.20, .02
w2 → w3	−.07	.04	.231	−.13, .03	−.06	.04	.266	−.12, .03	−.07	.04	.209	−.13, .03	−.06	.04	.286	−.12, .04
w3 → w4	−.02	.04	.697	−.10, .06	−.03	.04	.541	−.11, .06	−.02	.04	.655	−.10, .06	−.03	.04	.560	−.10, .06

Notes. All results are at the within-person level. Significant autoregressive coefficients indicate stability across waves and insignificant autoregressive coefficients indicate change. Results are considered significant at p < .0125 to correct for potential alpha inflation due to multiple testing. w1 = Wave 1 (May). w2 = Wave 2 (June). w3 = Wave 3 (July). w4 = Wave 4 (October).

Mulder, J. D., & Hamaker, E. L. (2021). Three extensions of the random intercept cross-lagged panel model. Structural Equation Modeling: A Multidisciplinary Journal, 28(4), 638–648. https://doi.org/10.1080/10705511.2020.1784738

 Web of Science ®Google Scholar

Data availability statement

The data that support the findings of this study are openly available on the Open Science Framework (OSF) under “Study 2” at https://osf.io/kxt2w/?view_only=5e9bc7f73ee8464cb7a26cf473f9ae1e. This OSF also includes other online supplementary files, including an overview of the statistical analyses and a description of the results of the additional analysis. This study was part of a larger project, which has been preregistered on the OSF at https://doi.org/10.17605/OSF.IO/R4FE5.