.1. Data sources
We linked data on MRSA incidence with patient reports of perceived hospital cleanliness, and health workers’ reports of availability of handwashing facilities for 126 Acute Trusts. Data on hospital-borne MRSA incidence per 100,000 hospital bed-days were taken from Public Health England’s annual reports (Public Health England, 2015). Data on patient-reported cleanliness were obtained from the Picker Institute NHS Patient Survey Programme (Care Quality Commission, 2010–2014) while data on handwashing facilities were from the Picker NHS National Staff Survey (Picker Institute Europe, 2010–2014). The two surveys are commissioned by NHS England from Picker Institute Europe. In the first, each Trust sends a questionnaire to 850 patients who have spent at least one night in the hospital between June and August each year. All the sampled patients are asked “In your opinion, how clean was the hospital room or ward (toilets and bathrooms) that you were (used) in? Very clean (excellent), fairly clean, not very clean, not clean at all”. In the NHS staff survey, each Trust selects a random group of staff (sample sizes will depend on the number of staff employed by the organisation from 600 to 850) to be interviewed. The survey asks all selected employees about their job, management, health/safety, and well-being in the Trust as well as their personal development. Here we are interested in a particular question “Are handwashing materials always available? Yes/No”. All data were for the years 2010–2014. Data on whether hospitals outsourced cleaning were obtained from Patient Environment Action Teams (2010-2)(Health & Social Care Information Centre, 2010–2014b) and Patient-Led Assessments of the Care Environment (2013-4) (Health & Social Information Centre, 2013–2014) (the name changed but collection practices did not). In practice, virtually all Trusts either fully outsourced or operated in-house cleaning services. Additional data on economic costs of cleaning per bed, staff numbers, patient mix and demographics, as well as size and services provided by the hospitals were taken from Estates Return Information Collection (ERIC) for the period 2010–2014 (Health & Social Care Information Centre, 2010-2014a). Table 1 in the web appendix provides further descriptive statistics for all variables used in the study.

Our initial sampling frame included all acute general hospital Trusts in England. We excluded single speciality orthopaedic, cardiac/ophthalmology/otolaryngology, gynaecology and paediatric hospitals given their atypical case mix (namely, Harefield, Royal National Orthopaedic, Royal National Throat, Nose and East Hospital, Papworth, Alder Hey, Robert Jones and Agnes Hunt Orthopaedic, Great Ormond, Moorefield Eye Hospital, Birmingham Children’s Hospital, Heart of England NHS Foundation, Birmingham women’s NHS foundation Trust and Sandwell and West Birmingham Hospital NHS Trust, and Royal Free Hampstead NHS Trust). Between 2010 and 2014 there were a total of 320 Acute Care Trusts, of which complete data existed for 201. It was not possible to track data over time in 119 Trusts because they changed identification codes during mergers. Of the 201, 140 report MRSA rates for the entire period. To avoid potential confounding from mixed service providers and switching (and numbers were too small to permit difference-in-difference analysis), we exclude a further four Trusts that use a combination of in-house and outsourced services and another four that changed from in-house to outsourcing (2) or vice-versa (2). Another four Trusts were removed because of small numbers or because they reported very high numbers (e.g. 7-fold higher than the median that indicated major outbreaks likely to have specific causes). Thus, our final analytical sample includes 126 acute Trusts. Of these 51 outsourced cleaning and 75 retained it in-house. Web appendix Fig. 1 further documents the sample inclusion criteria.

It is important to ascertain whether there were any pre-existing differences between hospitals that outsourced cleaning and those retaining it in-house, which might bias results, for example if hospitals with a worse cleaning record selectively outsourced it. Unfortunately, there are few sources of data that would allow such a comparison. One that does provide some insight is the dataset on hospital cleanliness, as assessed by the Healthcare Commission, from between three and five years prior to the data used in the main analysis, which start in 2010. We use these data to explore whether our results are consistent after adjusting for pre-existing differences in hospital sites, as measured by this indicator many years before the differences in out-sourcing (see web appendix Fig. 2 for more details).