What can we learn from Covid-19 pandemic's impact on human behaviour? The case of France's lockdown

The Covid-19 crisis led to unprecedented nationwide lockdowns to restrict human interactions. Behavioural measures (distancing, masking) were crucial pending vaccines or treatments, making public compliance a key public health objective. Prior work suggests compliance varies with gender, age, anxiety, trust in government, and risk perceptions. This study aimed to identify the individual characteristics and perceptions associated with respecting French Government lockdown measures during the first national lockdown. Using an online survey of 12,064 adults across metropolitan France conducted from March 24 to May 10, 2020, the authors investigated behaviour prior to lockdown and perceptions throughout, to inform future management strategies.

The introduction reviews research indicating: higher compliance among women, older adults, those with anxiety and higher government trust; risk perceptions (fear of virus, self-estimated vulnerability, perceived severity) drive protective behaviour; individuals often underestimate personal risk relative to others; government communication can modulate risk perception and increase self-protective behaviours; compliance is linked more to health beliefs and situational perceptions than to stable dispositions. Studies also note the impact of mobility restrictions on transmission and the potential of behavioural measures to reduce need for lockdowns.

Design: Cross-sectional online survey conducted from March 24, 2020 to May 10, 2020 (second week to last day of France’s first lockdown). Adults living in metropolitan France were targeted; no restrictions on age, gender, employment, or education. Recruitment via professional/personal networks, associations, political parties, companies, newspapers, with encouragement to share. The survey collected data on demography, occupations, mobility and life habits, social perceptions, usual residency and residency during lockdown, perceptions of lockdown, feelings, and willingness to change behaviour post-pandemic. Participants: Initial responses >12,000; exclusions included incomplete entries (not recorded), age <18 (n=153), and those not living in metropolitan France or residing abroad during lockdown (n=241). Final sample n=12,064. Data collection: LimeSurvey in week 1, then Google Forms to increase accessibility and remove participation limits. Participation was anonymous; informed statement presented before survey. Ethical standards were followed. Context: Key French Covid-19 dates and national measures from January–May 2020 are documented (e.g., progressive restrictions, national lockdown from March 17, extensions, and end on May 11). Statistical analysis: Conducted in R. Used chi-squared tests for 2x2 contingency tables, ANOVA for numerical variables, partial binomial multiple regressions for bivariate variables, and partial multinomial logistic regression for categorical variables; partial models included multiple predictors to control confounding. Cohen’s d reported and interpreted (very small to huge). Multiple comparisons corrected via Bonferroni; significance p≤0.05. Clustering: K-means clustering on perception items. Number of clusters determined by comparing solutions from 2 to 10 using 24 validity indices (Euclidean distance), selecting the median optimal number (NbClust package). The procedure iterated until convergence; details per Charrad et al., 2014. Optimal solution: 3 clusters. Text mining: Free-text responses to the prompt about lockdown impact (positive/negative/no change; “Could you develop?”) were cleaned (lowercasing, whitespace, removing numbers/punctuation/stopwords). Word frequencies computed per cluster; cluster-specific words identified via variation coefficient (VC). Top 20 high-VC, high-probability words per cluster considered specific.

Sample: N=12,064. Week-by-week comparisons showed unbalanced distributions across age (d=0.878), socio-professional category (d=0.865), region (d=0.560), education (d=0.504), home composition (d=0.439), gender (d=0.359), type of residency (d=0.312), and perceived salary (d=0.153). Behaviour before lockdown: 11.7% changed usual residence in the hours/days before lockdown. Reasons: to be near family/friends 76.7%; better living environment 59.6%; increase living area 48%; avoid over-population 20.1%; unexpected reason 4.7%. Distances: 13.9% same city; 14.9% same department; 19.7% same region; 51.4% farther. Moves tended to preserve residency type (X=35.7; p<0.0001; d=0.318); suburban and countryside residents mostly moved within same type; flat-dwellers represented 81.1% of those leaving their homes. Predictors of moving pre-lockdown (partial logistic regression): familial organisation (F(5,12020)=51.392; p<0.0001; d=0.298; higher for house-shares and those living alone without children); professional situation (F(10,12020)=12.98; p<0.0001; d=0.208; higher for students, retired, unemployed); age (F(1,12020)=118.034; p<0.0001; d=0.198; slope -0.044; younger more likely); residency type (F(2,12020)=52.152; p<0.0001; d=0.186; higher for city flats); perceived salary (F(4,12020)=6.954; p<0.0001; d=0.096; higher for very comfortable); gender (F(1,12020)=6.447; p=0.011; d=0.046; higher for women). Region (F(12,12020)=1.154; p=0.311; d=0.068) and education (F(6,12020)=1.894; p=0.078; d=0.061) not significant. Perception clusters (k-means, 3 clusters):

• Cluster 1 (30.8%): Positive adaptation and concern about the pandemic; highest acceptance and respect of measures; high worry for close relations; engaged in new activities.
• Cluster 2 (33.8%): Pessimistic, worried, stressed; felt loss of freedom; lowest agreement with Government measures; lower respect for rules; high negative affect.
• Cluster 3 (35.4%): Felt largely unaffected; high sense of safety; believed others respected measures; lower personal compliance. Temporal trends: Proportion of Cluster 1 decreased (F(1,12062)=45.449; p<0.0001; d=0.123); Cluster 2 increased (F(1,12062)=32.766; p<0.0001; d=0.104); Cluster 3 stable (F(1,12062)=0.618; p=0.43; d=0.014). Between-cluster differences (partial multinomial): perceived salary (d=0.284), gender (d=0.270), education (d=0.174), type of residency (d=0.129), home composition (d=0.128), region (d=0.123), socio-professional category (d=0.116), and pre-lockdown residence change (d=0.067). No age effect (d=0.083). Text mining: Cluster 1 specific words included “refocus”, “reading”, “yoga”, “cooking”, “best”, “essential” (high VCs). Cluster 2: “depression”, “anger”, “worry”, “boredom”, “fear”. Cluster 3: “noise”, “peaceful”, “break”, “read”, “gardening”, “spring”, “games”, “appreciate”. Intentions to change post-lockdown (partial multinomial): significant differences for “change the vision of others” (X=240.78; p<0.0001; d=0.282; slopes: C1 2.19%; C2 1.26%; C3 -3.45%); “get closer from relations” (X=214.38; p<0.0001; d=0.267; slopes: C1 1.87%; C2 0.78%; C3 -2.66%); “change my life habits” (X=48.95; p<0.0001; d=0.127; slopes: C1 1.36%; C2 0.05%; C3 1.40%); “change my eating habits” (X=28.77; p<0.0001; d=0.098; slopes: C1 0.95%; C2 -0.15%; C3 -0.79%). “Change my travel habits” not significant (X=5.42; p=0.066; d=0.042).

Pre-lockdown mobility: Over 10% of participants relocated just before lockdown, often over long distances, likely contributing to spatial spread, consistent with administrative and mobile phone data. Moves were more common among younger/older, those able to work from home, and non-family households, likely facilitated by the one-day announcement-to-enforcement delay. Perception profiles and compliance: Three profiles emerged. Cluster 1 showed highest acceptance and compliance, balancing concern about Covid-19 with positive reorientation of time and relationships. Its share decreased over time. Cluster 2 viewed lockdown as punitive, experienced high negative affect, low trust and satisfaction with measures, and showed lower compliance; this group grew over time, aligning with literature on quarantine-associated frustration, boredom, and reduced compliance. Cluster 3 also had lower compliance but due to low perceived personal and relational risk and belief that government measures and others’ compliance were sufficient; they felt largely unaffected by the virus or measures. Social inequality: Differences between Cluster 2 and Cluster 3 suggest the role of living environment and socioeconomic status. Cluster 2 over-represented residents of Île-de-France, unemployed, those perceiving insufficient purchasing power, living in city flats. Cluster 3 over-represented highly educated individuals with comfortable quality of life who often left cities before lockdown. These patterns reflect social and economic inequalities shaping compliance and risk exposure. Risk valuation and fear: Compliance appears to depend on the balance between perceived infection risk/safety and perceived adequacy of government measures. Fear can increase compliance when measures are seen as effective, but in contexts of uncertainty it may also promote risky decisions or reduce preventive intentions, especially with collectivist orientations. Notably, age was not associated with cluster membership, suggesting perceived exposure and social norms may override objective risk factors in shaping compliance. Implications: For groups with low trust and high fear in unfavourable environments, clear, trust-building communication and support are needed to maintain compliance. For groups with low perceived risk and reliance on others’ compliance, messaging should address personal relevance, cooperative norms, and collective responsibility.

Respect of lockdown measures was mainly altered by: (i) unfavourable living environments linked to social and economic inequities, combined with high fear and low trust in governmental measures; and (ii) low risk perception reinforced by the belief that others comply. Compliance seems more influenced by situational perceptions than by pre-existing dispositions. Governments should deliver clear, skilled communication to build trust, limit fear, and foster cooperative behaviours, potentially reducing the necessity for future lockdowns.

• Sample imbalance with over-representation of highly educated participants and executives; controlled statistically via partial models but may limit generalizability.
• No data from the first week of lockdown, likely the most anxiety-inducing and uncertain period.
• No recording of participants’ or close relations’ Covid-19 test status; the focus was on representations rather than medical data, and collecting such data would have required ethical approvals that could have precluded timely data collection.