Public and privately run prisons: how they compare

Taken together, the 15 privately run prisons do about as well as public prisons of the same kind. Their average position in their peer groups is a little higher (56 against 49 out of 100), but a gap that size could easily arise by chance. The one difference that holds up is fewer assaults on staff than expected, and that may partly reflect how incidents are recorded.

• Privately run prisons15run by G4S, Serco, Sodexo; 105 public sector
• Average position in group56 vs 49contracted vs public, out of 100 (top of group = 100)
• Clear differences1 of 10measures where chance is an unlikely explanation

Scores within peer groups

Every prison is placed within its own peer group, so a privately run local prison is only compared with public local prisons. The charts show how far the privately run prisons sit above (right) or below (left) public prisons on average, with a 95% interval. An interval that crosses the centre line means the difference could be chance. The hollow markers repeat the comparison using only prisons opened since 1990: all but one privately run prison is that new, so this separates the operator from the age of the buildings.

• All prisons
• Opened since 1990 only

Overall score+7 (−6 to +20)

Safety+4 (−11 to +17)

Decency+8 (−5 to +21)

Regime+14 (−1 to +27)

Resettlement−11 (−24 to +1)

Official HMPPS score+7 (−8 to +22)

Show as a table

Measure	All prisons	Opened since 1990 only
Overall score	+7 (−6 to +20; p 0.41; 56 vs 49)	+11 (−8 to +29; p 0.33; 55 vs 44)
Safety	+4 (−11 to +17; p 0.65; 53 vs 50)	+12 (−5 to +29; p 0.25; 53 vs 41)
Decency	+8 (−5 to +21; p 0.33; 57 vs 49)	+9 (−9 to +28; p 0.41; 57 vs 48)
Regime	+14 (−1 to +27; p 0.10; 62 vs 48)	+22 (+3 to +40; p 0.05; 60 vs 38)
Resettlement	−11 (−24 to +1; p 0.18; 40 vs 51)	−15 (−34 to +3; p 0.18; 39 vs 54)
Official HMPPS score	+7 (−8 to +22; p 0.41; 56 vs 49)	+2 (−16 to +20; p 0.85; 56 vs 55)

Incidents and reoffending against expectation

These compare each prison's rate with the rate expected for who it holds (the same expectation used on prison pages), then compare privately run with public prisons. Left of the centre line means fewer incidents than public prisons, relative to expectation.

• All prisons
• Opened since 1990 only

Prisoner-on-prisoner assaults−9% (−27% to +11%)

Assaults on staff−38% (−64% to −6%)

Self-harm incidents+20% (−5% to +49%)

Proven reoffending+12% (−5% to +34%)

Show as a table

Measure	All prisons	Opened since 1990 only
Prisoner-on-prisoner assaults	−9% (−27% to +11%; p 0.45)	−27% (−41% to −8%; p 0.04)
Assaults on staff	−38% (−64% to −6%; p 0.03)	−54% (−74% to −28%; p 0.00)
Self-harm incidents	+20% (−5% to +49%; p 0.35)	+7% (−17% to +39%; p 0.66)
Proven reoffending	+12% (−5% to +34%; p 0.29)	−8% (−20% to +5%; p 0.51)

What stands out

• Assaults on staff. Relative to expectation, privately run prisons record 38% fewer assaults on staff than public prisons (−64% to −6%). Among prisons opened since 1990 the gap is 54% fewer, and prisoner-on-prisoner assaults are also lower (−27%). This could mean safer prisons, or different recording: contracted prisons already report staff sickness through a different route from public prisons.
• Regime. Privately run prisons score higher on purposeful activity (inspection and survey grades, and the share of prisoners in purposeful activity): +14 points overall (could be chance) and +22 points against public prisons of the same era (only just clear of chance).
• Resettlement. They score lower on housing and employment after release (−11 points), though not clearly.
• Self-harm and reoffending are a little higher in privately run prisons relative to expectation, but neither difference is clear.
• The official HMPPS score shows the same picture as ours: +7 points overall and almost no gap among newer prisons (+2).

We tested 10 measures. By chance alone, about one in twenty tests will show a "clear" difference, so a single result near the line deserves caution.

By operator

Comparisons of single operators rest on four to six prisons each, so treat them as description, not verdict.

Every privately run prison

Band is the prison's position in its own peer group. Group tables can be filtered by operator.

What this cannot tell you

• Which is better value. Costs are not part of this comparison.
• Cause and effect. Prisons were not contracted out at random. Most privately run prisons were built in the last 30 years under private finance, for particular roles, so the comparison is between different estates, not the same prison under two kinds of operator.
• Small numbers. 15 privately run prisons, none of them open prisons and only one (Dovegate) in the category B and high security group. Two (Peterborough male and female) share one reporting unit and are left out of the incident comparisons.
• Changes of operator. Lowdham Grange was privately run until July 2024 and is counted as public, as HMPPS classifies it; leaving it out changes nothing. Altcourse moved from G4S to Sodexo in June 2023.
• Recording. Incident counts depend on what is recorded, and recording practice may differ between operators.

How the comparison is made

Positions run from 0 (bottom of the peer group) to 100 (top). Incident and reoffending outcomes use each prison's ratio of actual to expected, compared on a log scale. The 95% interval comes from resampling prisons within each peer group and operator type 2,000 times; the chance of a difference this large arising by chance comes from reshuffling which prisons are privately run within each peer group 5,000 times. Every figure is in the data download (v0_3_operator_comparison.csv and v0_3_operator_prisons.csv). Read the methodology.