normally do, without instructions on length of time or thoroughness. The volunteers allocated to handwashing were then provided with a paper towel to dry their hands. A wet NaCl-soaked charcoal swab was then wiped across the fingers of the dominant hand of the participant. The participants were finally given an alcohol gel to clean their hands (78% total alcohol content, Ethanol 71% / Propanol 29%, Softalind Viscorub, Braun-Melsungen). The swabs were returned to the laboratory within 5 hours of being taken. In total, 480 samples were collected; 160 after handwashing with plain soap, 160 after handwashing with water alone and 160 with no handwashing. During the experimental phase we measured the amount of time taken to conduct handwashing with and without soap, once for each volunteer. Upon arrival at the laboratory the swabs were immediately cut into a universal tube containing 10 mL of Purple MacConkey broth using aseptic techniques. The swabs were incubated at 35 °C for 48 hours. All samples were then streaked onto the MacConkey agar No.3 and Bile Aesculin agar. MacConkey agar No. 3 is a selective media which can differentiate between coliforms and non-lactose fermenters, whilst inhibiting gram-positive cocci. These plates were incubated for 18–24 hours at 35 °C. For all other colonies produced on MacConkey agar No. 3 and those which were spot indole negative, a gram stain, catalase and oxidase test was carried out followed by an API 20E biochemical test to determine the identity of the bacteria. Bile Aesculin agar is a differential medium for the isolation of Enterococcus spp. and group D Streptococcus and inhibition of other gram positive bacteria. These plates were incubated at 37 °C for 18–24 hours. Enterococcus and Group D Streptococcus spp. are able to hydrolyse the aesculin to form aesculetin, producing a brown/black complex. Any white colonies on Bile Aesculin agar were presumed to be Staphylococcus spp. and any black colonies were tested with Lancefield group D antisera. Agglutination indicated a positive result for Enterococcus spp. The prevalence of bacterial contamination in the three study arms (soap, water, no handwash) was compared using logistic regression. Since the same volunteers repeatedly underwent testing, within-subject correlation was accounted for by the use of generalised estimating equations (GEE) with robust standard errors. If the cell numbers were too low for conducting regression analysis, Fishers exact test was used instead, ignoring clustering (the design effect was found to be low, see results).