Table 3. A covariate balance check for falls on compliant versus regular flooring.

Variable (measure, m = mean, p = proportion)	Falls on compliant flooring	Falls on regular flooring	Difference (compliant-regular)	P-value for difference
Age (m)	85.0	82.8	2.17	0.225
Women (p)	0.68	0.75	-0.07	0.686
BMI (m)	23.8	24.4	-0.60	0.526
Visual impairment (p)	0.29	0.31	-0.02	0.930
Cognitive impairment (p)	0.73	0.67	0.06	0.773
Hip protector (p)	0.18	0.14	0.04	0.750
Walking ability (p)
Safe walker	0.34	0.49	-0.15	0.548
Unsafe walker	0.63	0.45	0.18	0.445
Non-walker	0.04	0.07	-0.03	0.498
Location (p)
Bathroom	0.00	0.19	-0.19	N/Aa
Bedroom	0.81	0.69	0.12	0.328
Communal area	0.19	0.12	0.07	0.431
Activity (p)
Low transfer (e.g. bed to wheelchair)	0.41	0.29	0.12	0.064
Sitting or lying down	0.13	0.10	0.03	0.374
Standing or walking	0.33	0.39	-0.06	0.448
Unknown	0.13	0.22	-0.09	0.002
Time of day (p)
6-10AM	0.13	0.12	0.01	0.810
10AM-2PM	0.13	0.15	-0.02	0.610
2-6PM	0.19	0.22	-0.02	0.632
6-10PM	0.17	0.21	-0.03	0.481
10PM-02AM	0.20	0.16	0.04	0.547
2-6AM	0.17	0.14	0.03	0.427

Models were fitted separately for each variable. Categorical variables were transformed to dummy variables for each category prior to estimation. Continuous variables were analyzed using linear regression, and binary variables (presented as proportions) were analyzed using logistic regression (differences from these models are presented as marginal means obtained using the margins command in Stata 12). All models were fitted using clustered standard errors to account for within-individual correlation.

^aA logistic model could not be fitted to the data, because the Kradal flooring was not installed in bathrooms.