Guides / impact measurement
Logic models and social return on investment
A logic model sets out how an activity is supposed to produce a result. SROI puts a money value on that result and divides by what it cost. One is a planning tool that costs nothing and prevents most evaluation problems. The other produces a single memorable ratio that is far more sensitive to your assumptions than to your programme. This guide covers both, with the arithmetic shown.
A logic model is a one-page chain from inputs to activities to outputs to outcomes to impact. It forces you to state what you expect to change and how, which is what makes an evaluation possible later. SROI takes the outcomes from that chain, assigns each a financial proxy, subtracts what would have happened anyway, discounts future years, and divides the total by the investment. The output is a ratio such as 9:1. The logic model is the prerequisite; without one, SROI is valuing outcomes nobody agreed to in advance.
Running 40,000 draws, with deadweight uniform between 15% and 50%, attribution between 10% and 35%, displacement between 5% and 20%, drop-off between 15% and 45% and the proxy between £9,000 and £15,000:
median 7.84 : 1
5th pct 5.08 : 1
95th pct 11.77 : 1
P(SROI > 5 : 1) = 96%
P(SROI > 10 : 1) = 18%
"We are 96% confident the return exceeds 5:1, under stated assumption ranges" is a claim that survives questioning. "9.31:1" does not, and the second statement is the one that gets printed.
The structure
The chain, and the step everyone skips
| Stage | What it is | Example |
|---|---|---|
| Inputs | What you put in | £120,000, four staff, premises |
| Activities | What you do | Weekly employability sessions, one-to-one coaching |
| Outputs | What you produce, counted | 240 people attended, 1,100 sessions delivered |
| Outcomes | What changes for them | 85 people in work six months later |
| Impact | The change you caused | The portion of those 85 attributable to you |
The skipped step is between outcomes and impact. Most reports stop at outcomes and present them as impact. The 85 people in work is an outcome. How many of those 85 would have found work without you is the question that turns it into impact, and it is the question the whole discipline exists to answer.
A good logic model has a column most templates omit: what has to be true for each arrow to hold. "Attendance leads to improved confidence" assumes the sessions address confidence, that people attend enough of them, and that confidence is what was limiting them. Writing those down turns a diagram into something testable, and it is where an evaluator will start.
The most common confusion
Outputs are not outcomes
Funders ask for outcomes and receive outputs, reliably and almost universally.
| Output | Corresponding outcome |
|---|---|
| 240 people attended a course | 85 are in work six months later |
| 1,200 leaflets distributed | Screening uptake rose in the target group |
| 50 staff trained | Practice changed in the way the training intended |
| A helpline answered 3,000 calls | Callers resolved the problem they rang about |
The left column is within your control and easy to count. The right requires following people after they leave, which costs money and sometimes produces unwelcome answers. That asymmetry is why output reporting persists.
The practical test: if the number would be the same whether or not the programme worked, it is an output.
Measurement
Choosing indicators that survive scrutiny
Monetisation
What SROI adds, and what it costs you
SROI converts outcomes into money so that unlike things can be added together and compared with the investment. The appeal is obvious: a single number any funder understands.
Four adjustments stand between the raw outcome and the credited impact, and they are the substance of the method.
| Adjustment | Question it answers | Typical range |
|---|---|---|
| Deadweight | How much would have happened anyway? | 20–50% |
| Attribution | How much is down to other organisations? | 10–40% |
| Displacement | Did the benefit simply move from elsewhere? | 0–20% |
| Drop-off | How fast does the benefit fade? | 10–50% per year |
Each is an estimate. None is observable. Together they usually remove half the headline value, and the figure you choose for each one moves the final ratio more than anything your programme actually did.
Worked example one
Calculating an SROI ratio
An employability programme costing £120,000 supports 85 people into work. The financial proxy is £12,000 per person per year, combining benefits no longer claimed and additional earnings.
Step 1. Gross outcome value
85 × £12,000 = £1,020,000 in year one
Step 2. Apply the three reducing adjustments
Deadweight 30%, attribution 20%, displacement 10%. They multiply rather than add.
(1 − 0.30) × (1 − 0.20) × (1 − 0.10) = 0.504
Only 50.4% of the gross value is credited to the programme. Year one impact is £514,080.
Step 3. Project forward with drop-off
The benefit fades at 25% a year over a three year horizon.
Step 4. Discount to present value
At the Green Book rate of 3.5%.
| Year | Impact | Discount factor | Present value |
|---|---|---|---|
| 1 | £514,080 | 0.9662 | £496,696 |
| 2 | £385,560 | 0.9335 | £359,924 |
| 3 | £289,170 | 0.9019 | £260,815 |
| Total present value | £1,117,435 | ||
Step 5. The ratio
SROI = £1,117,435 / £120,000 = 9.31 : 1
£9.31 of social value for every £1 invested. That is the number that will appear in the annual report, and on its own it is close to meaningless. The next section shows why.
Worked example two
The same programme, three ratios
Keep the cost, the 85 people and the £12,000 proxy exactly as they are. Change only the three judgement calls.
| Assumption set | Deadweight | Attribution | Displacement | SROI |
|---|---|---|---|---|
| Optimistic | 15% | 10% | 5% | 13.43 : 1 |
| Middle | 30% | 20% | 10% | 9.31 : 1 |
| Cautious | 50% | 35% | 20% | 4.80 : 1 |
Identical programme, identical outcomes, identical cost. The ratio ranges by a factor of nearly three, and every one of those assumption sets is defensible.
The horizon matters more than the judgement calls
The table above varies deadweight, attribution and displacement. Those are the three adjustments SROI guidance tells you to think about, so they get the attention. Two other choices move the number further and are usually made without comment.
| Variation from the published case | Present value | SROI |
|---|---|---|
| 1 year only | £496,696 | 4.14 : 1 |
| 2 years | £856,620 | 7.14 : 1 |
| 3 years, as published | £1,117,435 | 9.31 : 1 |
| 5 years | £1,443,385 | 12.03 : 1 |
| 10 years | £1,731,779 | 14.43 : 1 |
Changing nothing but the horizon moves the ratio from 4.14 to 14.43, a wider span than the whole optimistic-to-cautious range in the previous table. Nobody audits a horizon, because it looks like a technical parameter rather than a judgement.
The discount rate, by contrast, barely matters over a short horizon.
no discounting: £1,188,810 → 9.91 : 1
3.5% Green Book: £1,117,435 → 9.31 : 1
6%: £1,070,921 → 8.92 : 1
A full percentage point of discount rate is worth less than four months of horizon. Arguing about the discount rate while leaving the horizon unstated is the wrong way round, and it happens often.
An SROI ratio is not a measurement of a programme. It is a measurement of a programme plus a set of judgements, and the judgements carry more weight than the evidence. Two charities doing identical work can honestly report 5:1 and 13:1. Comparing published ratios between organisations is therefore meaningless unless both used the same assumptions, which they almost never do.
The remedy is not to abandon the method. It is to publish the sensitivity range rather than the point estimate, and to state every assumption with its source. An SROI reported as "between 4.8:1 and 13.4:1 depending on attribution assumptions, with 9.3:1 as our central estimate" is honest and still useful. A bare 9.3:1 is neither.
The weakest link
Financial proxies
Every outcome has to become a pound figure before it can enter the calculation. For some that is straightforward. For most it is not.
| Outcome | Proxy approach | Confidence |
|---|---|---|
| Moved into employment | Benefits saved plus earnings gained | Reasonable, both observable |
| Avoided a hospital admission | National reference cost per admission | Reasonable, published |
| Improved wellbeing | Wellbeing valuation from national datasets | Contested, wide range |
| Increased confidence | Cost of a comparable course, or stated preference | Weak, largely arbitrary |
| Stronger community ties | Various, none satisfactory | Very weak |
The pattern is unavoidable: the outcomes that matter most to the people served are the hardest to value, and the values assigned to them are the least defensible. A programme whose main benefit is confidence or connection will have an SROI built mostly on contested proxies.
How much the proxy itself is worth
The published example values moving into employment at £12,000 a person. That figure is a choice, and it scales the whole result linearly.
| Proxy per person | Present value | SROI |
|---|---|---|
| £9,000 | £838,076 | 6.98 : 1 |
| £10,500 | £977,755 | 8.15 : 1 |
| £12,000 | £1,117,435 | 9.31 : 1 |
| £13,500 | £1,257,114 | 10.48 : 1 |
| £15,000 | £1,396,794 | 11.64 : 1 |
A quarter either way on one proxy moves the headline by more than two points. Because the relationship is linear, the ratio is only ever as defensible as the proxy, which is the argument for citing the source of every proxy and reporting the range the source itself gives.
One figure worth computing and almost never published: the value at which the programme stops washing its face.
SROI reaches 1 : 1 at a proxy of about £1,290 per person
The programme would have to be worth less than a ninth of the assumed figure before it stopped returning more than it cost. That is a far more robust claim than 9.31 : 1, and for most audiences it is the more persuasive one.
The HM Treasury Green Book and the Social Value Bank are the usual UK sources. Using a published proxy and citing it is far stronger than deriving your own, even when the published figure is imperfect, because it is transparent and comparable.
Doing it
Doing it in practice
Spreadsheet, and why that is the right answer
SROI is arithmetic, not statistics. A spreadsheet with one row per outcome and columns for quantity, proxy, deadweight, attribution, displacement, drop-off and discount factor is the standard approach, and it has a real advantage: every assumption is visible in a cell that a sceptical reader can change.
Build the sensitivity analysis into the same sheet. Three columns of assumptions, three ratios, no extra work.
R, for the parts that are statistical
The outcome measurement underneath SROI is where statistics belongs.
t.test(post, pre, paired = TRUE) # did the outcome change
DescTools::BinomCI(85, 240, method = "wilson") # proportion into work, with CI
MatchIt::matchit(treated ~ age + prior_benefit, data = d) # a comparison group
That last one matters most. Deadweight is usually guessed; matching against a comparison group estimates it instead. If you can obtain any comparison data at all, the single largest source of uncertainty in the whole model becomes measurable.
Monte Carlo, if the audience is sophisticated
Rather than three scenarios, draw each assumption from a plausible distribution a few thousand times and report the resulting range of ratios. Fifty lines of R. It turns the sensitivity question from three hand-picked cases into a proper distribution, and it is increasingly expected by larger funders.
The spreadsheet should contain every assumption, its source and its rationale, and should be available on request. The published report should contain the central estimate, the range, and the three or four assumptions that drive it. Mixing the two produces a report nobody reads and a model nobody can audit.
Avoid these
Six mistakes that cost you credibility
| Mistake | Why it matters |
|---|---|
| Reporting outputs as outcomes | 240 people attended is not 85 people in work. If the number would be identical whether or not the programme worked, it is an output. |
| Publishing a single ratio | The same programme yields 4.8:1 or 13.4:1 on defensible assumptions. A point estimate implies precision the method cannot deliver. |
| Guessing deadweight | It is usually the largest adjustment and the one most often picked to suit. A comparison group turns it from a guess into an estimate. |
| Comparing ratios between organisations | Different proxies, different assumptions, different horizons. Two identical programmes can differ threefold. |
| No baseline | Without a before measurement there is no change to value. This cannot be fixed retrospectively. |
| Extending the horizon to improve the ratio | Five years instead of three raises the number and lowers the credibility. Justify the horizon from the outcome, not the result. |
An SROI ratio is a programme plus a set of judgement calls, and the horizon alone can move it from 4:1 to 14:1. Send the model and you get a sensitivity range rather than a point estimate, with every assumption sourced.
See statistical consultancy
Elaine HalliburtonProfessor of Applied StatisticsWriting it up
How to report it
The sentence that does the work: "Our central estimate is £9.31 of social value per £1 invested. Under more cautious assumptions about deadweight and attribution the figure is £4.80, and under optimistic assumptions £13.43. The range reflects judgement about what would have happened anyway, which we have not been able to measure directly."
Common questions
Frequently asked questions
What is a logic model?
A one-page chain showing how inputs become activities, activities become outputs, outputs become outcomes and outcomes become impact. Its value is that it forces you to state in advance what you expect to change and why, which is what makes evaluation possible later. The best versions add a column recording what must be true for each step to hold.
What is the difference between outputs and outcomes?
Outputs are what you produced and can count, such as 240 attendances. Outcomes are what changed for people, such as 85 in work six months later. The practical test is whether the number would be the same regardless of whether the programme worked. If it would, it is an output.
How do you calculate SROI?
Multiply the outcome quantity by a financial proxy, reduce it for deadweight, attribution and displacement, project forward with an annual drop-off, discount each year to present value, sum them, and divide by the investment. In the worked example this gave £1,117,435 of present value against £120,000 invested, a ratio of 9.31 to 1."
What is deadweight?
The share of the outcome that would have occurred without your programme. It is typically the largest adjustment, often 20% to 50%, and the hardest to establish. Guessing it is normal practice and the main weakness of most SROI studies; a matched comparison group turns the guess into an estimate.
Can I compare my SROI with another organisation's?
No, unless both used identical proxies, adjustments and time horizons, which is very rare. The same programme with the same outcomes and cost produces ratios between 4.8:1 and 13.4:1 depending on defensible assumption choices, so a comparison between organisations mostly compares their assumptions.
What discount rate should I use?
3.5% is the HM Treasury Green Book rate for UK public sector appraisal and is the conventional choice. State it explicitly, along with the time horizon, since both materially affect the result and a longer horizon always flatters it.
Is SROI worth doing?
It is worth doing properly or not at all. Done with a logic model, measured outcomes, a baseline, published proxies and an honest sensitivity range, it is a useful way to compare options within one organisation. Done as a single ratio derived from estimated adjustments, it is a marketing number, and funders increasingly recognise the difference.
Related services
If the findings go to a board or a funder
NHS data analysis
Audit, service evaluation and outcomes work for health organisations.
Fixed quoteStatistical consultancy
Design, analysis and reporting for charities, universities, the NHS and business.
From £950Independent statistician
A named statistician on your bid or an impartial check of analysis done in house.
Fixed quoteEvery piece of work is done and signed by a named statistician. Send the data or describe the project and you get a price and a deadline, usually within one working day.