What Is Technical Efficiency and How Is It Measured?

Technical efficiency asks a simple question about any farm, hospital, bank, or factory: is it getting as much as it could out of what it puts in? A technically efficient producer wastes nothing, so it can’t produce more without using more labour, land, money, or equipment. Nobody knows the true maximum in advance, so it is measured by comparing a producer with the best performers among similar ones, using data envelopment analysis (DEA) or stochastic frontier analysis (SFA). That makes a score relative: it says how far a unit trails its best peers, not how far it trails perfection.

The Idea Behind Technical Efficiency

Economists describe production as an ideal: the maximum output attainable from a given set of inputs.1Handbook of Applied Econometrics Volume II: Microeconomics. Handbook of Applied Econometrics Volume II: Microeconomics Trace that maximum across every possible input level and you get a boundary called the frontier. Real producers can sit on that boundary or below it, but never above it.2Journal of Econometrics. What Is Technical Efficiency and How Is It Measured? The distance below the frontier is the measure of technical inefficiency.

The best-practice line is only as good as the sample behind it: a line built from ten similar farms says less than one built from several hundred. The group also has to be genuinely alike in what it does, such as dairy farms with dairy farms, because a score against the wrong peers is meaningless.

Before any score exists, the analyst decides what counts as an input (staff hours, land, beds, capital) and what counts as an output (litres, patients treated, loans made). Leaving out an important input makes a unit look more efficient than it is, while leaving out an output makes a unit look wasteful when it was actually producing something else. Two studies of the same hospitals can reach different scores simply because they defined inputs and outputs differently.

A score is usually expressed between 0 and 1, where 1 means the producer is on the best-practice line. A farm scoring 0.8 could, in principle, grow about 25% more with the same land, seed, and labour, since 1 divided by 0.8 is 1.25. The same idea can be turned around to ask how much less input could produce today’s output.

The score also leaves a lot unjudged. It ignores prices, so a technically efficient producer can still be losing money. It ignores whether the output is wanted or well made, and it says nothing about whether the producer is the right size.

How It Fits With Allocative and Overall Efficiency

Since the economist Michael Farrell’s work in the 1950s, overall economic efficiency has been split into technical efficiency and allocative efficiency.3Journal of Optimization Theory and Applications. Decomposing Economic Efficiency into Technical and Allocative Components: An Essential Property Allocative efficiency asks whether a producer uses the right mix of inputs for their prices, such as machines versus workers. Overall efficiency is the two multiplied together, so weakness in either one pulls it down.4PubMed Central. What are the Technical and Allocative Efficiencies of Public Health Centres in Ghana? Allocative efficiency is usually worked out as what is left once technical efficiency and costs are known.3Journal of Optimization Theory and Applications. Decomposing Economic Efficiency into Technical and Allocative Components: An Essential Property

Among the input mixes on the best-practice line, allocative efficiency asks whether the producer chose the one that costs least at current prices. A producer can be fully technically efficient and still pay too much, for example by using many workers when machines have become cheap. Because this part depends on prices, it can change when wages or equipment costs change even if nothing about the producer’s practice has changed.

Say a clinic scores 0.9 technically and 0.8 allocatively. Its overall efficiency is 0.72, so it loses more to its input mix than to wasted effort. The fixes differ too: retraining staff addresses the first gap, while rebalancing staff and supplies addresses the second.

In practice the combined score can reveal widespread room for improvement. In one study of public health centres in Ghana, over 90% were inefficient once both parts were combined.4PubMed Central. What are the Technical and Allocative Efficiencies of Public Health Centres in Ghana? A facility that looks fine on technical efficiency alone can still waste money through a poor input mix. Many reports show only technical efficiency because it needs only quantities of inputs and outputs, not prices, and price data are often missing or distorted in public services.

Scale Efficiency and Whether Size Matters

If doubling every input always doubles output, economists call that constant returns to scale.5Quantitative Methods in Economics. EFFICIENCY AND RETURNS TO SCALE – A CONCEPT OF USING DETERMINISTIC APPROACH But if larger or smaller producers have natural advantages, returns to scale are variable, which gives a second frontier.5Quantitative Methods in Economics. EFFICIENCY AND RETURNS TO SCALE – A CONCEPT OF USING DETERMINISTIC APPROACH The score against the constant frontier divided by the score against the variable frontier gives scale efficiency.5Quantitative Methods in Economics. EFFICIENCY AND RETURNS TO SCALE – A CONCEPT OF USING DETERMINISTIC APPROACH A low scale efficiency points to the wrong size, not poor management.

A producer that is too small may be unable to spread fixed costs, such as a tractor or a scanner, over enough output. One that is too large may suffer from coordination problems, slower decisions, or stretched supervision.

Reading the two scores side by side makes this concrete. A small farm might score 0.7 under constant returns and 0.95 under variable returns. It is well run for its size, and its scale efficiency is about 0.74. The likely remedy is growing, merging, or sharing equipment.

The choice of assumption matters. Variable returns compares units mostly with others of similar size, which usually raises scores. Reports should say which assumption was used before anyone compares scores.

Measuring It With Data Envelopment Analysis

Data envelopment analysis, or DEA, is one of the most widely used measurement methods.6Wikipedia. Data envelopment analysis It compares the farms, hospitals, or banks being studied, which it calls units.6Wikipedia. Data envelopment analysis The best-performing units form the best-practice line, and every unit is scored against it, which is where the name “envelopment” comes from.6Wikipedia. Data envelopment analysis At least one unit always scores 1, even if all are wasteful by some outside standard.

For each unit, DEA solves its own optimisation problem. It looks for the most favourable weighting of that unit’s inputs and outputs that still keeps every unit’s score at or below 1. The method doesn’t impose one fixed price list on everyone, which is why units can be judged strong in different ways. The software runs this calculation once per unit, so a table of results lists one score per row.

That table gives each unit a score between 0 and 1. Inefficient units are matched to efficient peers they should learn from, and tables often show targets, meaning how much input could be cut.

Its appeal is flexibility. DEA needs few assumptions about how inputs turn into outputs.6Wikipedia. Data envelopment analysis It can handle several inputs and outputs at once, such as beds, staff, patients treated, and surgeries,6Wikipedia. Data envelopment analysis and it is quick to compute with standard optimisation software.6Wikipedia. Data envelopment analysis The catch is that it treats every gap from the best-practice line as inefficiency, including bad luck or data errors.

A score of 1 means best in this sample, not perfect. Adding better units to the sample can lower existing scores, and scores from different studies can’t be compared directly.

Measuring It With Stochastic Frontier Analysis

Stochastic frontier analysis, or SFA, starts by assuming a formula for how inputs produce output.2Journal of Econometrics. What Is Technical Efficiency and How Is It Measured? It then splits the gap between actual and predicted output into two parts: random noise and technical inefficiency.7OhioLINK Electronic Theses and Dissertations Center. Estimation of Technical Efficiency in Stochastic Frontier Analysis Noise can push output up or down, like weather or measurement error, while inefficiency can only push it down.8Wikipedia. Stochastic frontier analysis Because of this, a producer’s frontier can shift with its own luck, so a bad year isn’t counted entirely as inefficiency.7OhioLINK Electronic Theses and Dissertations Center. Estimation of Technical Efficiency in Stochastic Frontier Analysis If a drought cuts a farm’s harvest, DEA would record all of that as inefficiency, while SFA treats part of it as bad luck.

The price is having to choose the formula. The analyst picks a mathematical form linking inputs to output, with common choices being ones that let each input’s effect be estimated from the data. Statistical software then estimates the formula’s parameters from the full group of producers, rather than tracing the outer edge of the data. Unlike DEA, this gives standard errors and tests, so the reader can see how uncertain the estimates are. But because the formula must be chosen in advance, a poor choice can distort the scores. SFA also usually handles only one output at a time, whereas DEA handles several easily.

Choosing Between DEA and SFA

When the assumed formula is close to how production really works, SFA tends to measure efficiency more accurately than DEA.9Journal of Econometrics. Finite sample evidence on the performance of stochastic frontiers and data envelopment analysis using panel data As the formula gets further from reality, DEA becomes the better choice.9Journal of Econometrics. Finite sample evidence on the performance of stochastic frontiers and data envelopment analysis using panel data DEA also looks better when inefficiency is tied to the amount of inputs used.9Journal of Econometrics. Finite sample evidence on the performance of stochastic frontiers and data envelopment analysis using panel data

That gives a rough way to read a study. If the sector’s production process is well understood and the data are noisy with outside shocks, a statistical formula is more likely to be a sound approximation, which favours SFA. If the technology is unclear or producers make several outputs at once, DEA avoids the risk of imposing a wrong formula. The authors’ reasoning for their choice is more informative than the choice itself, and it is worth checking whether they tested their assumptions.

Both methods share a hidden trap. Managers often choose inputs knowing things the analyst can’t see, which can bias SFA efficiency scores.10Energy Economics. Dealing with endogeneity in stochastic frontier models: A comparative assessment of estimators DEA copes with some of this, but can be badly biased when an input rises strongly alongside true efficiency.11European Journal of Operational Research. Testing the accuracy of DEA estimates under endogeneity through a Monte Carlo simulation Many careful studies therefore run both methods and check whether rankings agree.

Agreement doesn’t prove the scores are right, because both can share the same bias from data or omitted inputs. If the methods disagree strongly, that is useful information: it suggests the formula or the noise assumptions are driving the results.

Where Technical Efficiency Is Used

In agriculture, scores are used to find what makes farms more efficient. Across a decade of farm studies worldwide, larger farm size, advice from extension services, access to credit, and education raised efficiency.12Journal of Agriculture and Food Research. Technical efficiency in agriculture: A decade-long meta-analysis of global research Subsidies had the biggest negative effect.12Journal of Agriculture and Food Research. Technical efficiency in agriculture: A decade-long meta-analysis of global research Payments that don’t depend on output can reduce the pressure to farm well, since income arrives either way.

Among smallholders in Ethiopia, land fragmentation, land quality, family size, and off-farm work also mattered.13Economies. Technical Efficiency of Smallholder Agriculture in Developing Countries: The Case of Ethiopia For a small farmer, advice and credit are things policy can change, unlike farm size or location. Extension services help by spreading practices from the best-performing farms to others, the same logic DEA uses when it points to peers. Fragmented land costs time and effort moving between plots, which a score can pick up as inefficiency even when the farmer works hard.

In a study of Virginia hospitals, the technically efficient ones also performed well on quality.14PubMed. Data envelopment analysis comparison of hospital efficiency and quality Efficiency gains need not come at patients’ expense, though quality should be measured alongside.14PubMed. Data envelopment analysis comparison of hospital efficiency and quality For banks, whether deposits count as an input or an output changes the scores so much that results can’t be compared.15Springer Proceedings in Business and Economics. The Production or Intermediation Approach?: It Matters Most bank studies treat deposits as inputs, so check which view a study used.16Review of Managerial Science. Technical efficiency in banks: a review of methods, recent innovations and future research agenda

Regulators use efficiency benchmarking to make evidence-based decisions about utilities such as water companies.17Utilities Policy. Benchmarking technical efficiency of water utilities in Chile under heterogeneity: A latent class frontier approach Germany’s energy regulator scores each network operator before setting revenue caps, and inefficiencies must be fully removed within one regulatory period.18Bundesnetzagentur. Efficiency benchmarking A low score means the operator’s allowed revenue is set as if it were already efficient. It must cut costs to stay profitable, and customers benefit from lower prices. Companies may dispute their peer group, since the score depends on who they are compared with.

Newer Twists on Measuring Efficiency

Newer DEA models count pollution and other undesirable outputs, so producers aren’t rewarded for dirtier output.19The Stata Journal: Promoting communications on statistics and Stata. Measuring technical efficiency and total factor productivity change with undesirable outputs in Stata In standard models more output of any kind earns a better score, which would reward a factory for producing more smoke. Undesirable-output models instead look for ways to expand good output while shrinking the bad, so a study can compare producers on economic and environmental performance in one score. One study of the European Union from 2000 to 2018 measured efficiency with emissions included.20Energies. Technical Efficiency and Productivity Change in the European Union with Undesirable Output Considered Counting pollution changes rankings: a factory producing more but polluting heavily may drop, while cleaner firms can look efficient despite lower output.

The Malmquist index tracks productivity change over time, separating producers catching up from the frontier itself moving.20Energies. Technical Efficiency and Productivity Change in the European Union with Undesirable Output Considered A firm’s productivity can rise because it closes the gap to the best-practice line, or because the line moves outward with new technology. The split tells a manager whether improvement came from better use of existing methods or from an industry-wide advance. A firm that improves but falls further behind a fast-moving frontier will show catch-up as negative. Recent work splits that frontier shift into change shared by all and how each firm experiences it.21Economic Modelling. Global and local technical changes: A new decomposition of the Malmquist productivity index using virtual units

Machine learning addresses a different limit. DEA can score existing units but can’t predict new ones and struggles with very large datasets.22Synergizing Data Envelopment Analysis and Machine Learning for Performance Optimization in Healthcare. Artificial Intelligence (AI) and Data Envelopment Analysis (DEA) in Healthcare Researchers therefore train machine learning models on DEA results to predict how new units will perform.23Mathematics. DEA and Machine Learning for Performance Prediction