Best Approaches to Evaluation: From Logical Frameworks to Bennett’s Hierarchy

Every public health program – from a district-level immunization drive to a national family planning mission – eventually faces the same question: did it actually work? Answering that question is the job of evaluation, and how you answer it depends entirely on the framework you use. Two of the most widely respected approaches in population and family health studies are the Logical Framework Approach (LFA) and Bennett’s Hierarchy. Both trace a program’s journey from resources to results, but they do it in different ways and reveal different kinds of insight. Understanding them is essential for anyone planning, monitoring, or studying social and health interventions.

Why evaluation needs a structured approach 🔗

Health and family welfare programs are complex. They involve money, people, supplies, training, behaviour change, and long-term shifts in community outcomes. Without a structured framework, evaluators end up reporting only what is easy to count – number of camps held, pamphlets distributed, or workers trained – while ignoring whether anything actually changed for the people the program was meant to serve. Structured approaches force evaluators to think systematically about the chain of cause and effect: what went in, what came out, and what difference it made.

This is where evaluation frameworks earn their place. They organise messy real-world programs into a logical sequence so that planners, funders, and communities can ask sharper questions and get more honest answers.

The Logical Framework Approach (LFA) 🔗

The Logical Framework Approach was developed in the late 1960s for the United States Agency for International Development (USAID) and has since become the global standard for designing and evaluating development projects. As Better Evaluation describes it, LFA is a systematic, visual approach that pushes users to consider the relationships between available resources, planned activities, and desired results. Its final product is a single-page matrix called the logframe.

The four levels of the logframe 🔗

The LFA matrix typically arranges program logic into four hierarchical levels:

Inputs: The resources invested in the program – money, staff, vehicles, contraceptives, training materials, time, and infrastructure. For a maternal health scheme, inputs would include funds released to the state, ANMs deployed, IFA tablets procured, and ambulances allocated.

Activities: The actions the program actually carries out using those inputs. This is what staff on the ground do every day – antenatal check-up camps, counselling sessions, home visits by ASHAs, or community awareness drives.

Outputs: The direct, tangible products of activities. According to the European Commission’s guidance on LFA, outputs are the goods and services delivered by the intervention. Examples: number of pregnant women registered, number of institutional deliveries conducted, or number of villages covered by mobile health units.

Outcomes and Impact: Outcomes are the changes in people or systems that result from outputs – increased contraceptive use, improved immunisation coverage, or higher health-seeking behaviour. Impact refers to the long-term, broader changes such as reduced maternal mortality, lower total fertility rate, or improved child health indicators.

The role of indicators and assumptions 🔗

What separates LFA from a simple to-do list is its insistence on two extra columns in the matrix: objectively verifiable indicators, means of verification, and assumptions or risks. Each level must specify how progress will be measured (for example, percentage of fully immunised children measured through NFHS data) and what external conditions need to hold true for the logic to work – political stability, monsoon patterns, community participation, or supply-chain reliability.

This forces planners to think hard about risk. A family planning program may distribute contraceptives flawlessly (output achieved), yet fail to reduce unintended pregnancies if a hidden assumption – say, that male partners will support contraceptive use – turns out to be wrong. The logframe makes such assumptions explicit instead of hiding them inside good intentions.

Strengths and limitations of LFA 🔗

The LFA has endured for sixty years because it produces clarity. It compels stakeholders to agree on objectives, indicators, and risks before a single rupee is spent. It also gives donors a clean way to compare projects and gives evaluators a roadmap to measure progress against.

Its limitations are equally well-known. As analysts have pointed out, logframes are often treated as compliance forms rather than living learning tools. Teams routinely confuse outputs with outcomes, and the rigid matrix can struggle with experimental or rapidly evolving programs where assumptions must be updated continuously. Done well, LFA is a thinking discipline; done poorly, it becomes a tick-box exercise.

Bennett’s Hierarchy 🔗

While LFA emerged from international development, Bennett’s Hierarchy came from a different world entirely – the United States Department of Agriculture’s extension service in the 1970s. Dr. Claude Bennett wanted a way to show that extension education programs were actually changing farmers’ lives, not just running events. The result was a “seven-link chain of events,” arranged as a staircase, that has since been adopted widely in agricultural extension, public health education, nutrition programs, and family welfare evaluation.

Bennett’s central insight is elegant: the higher up the staircase you can gather evidence, the stronger your claim of program impact. But evidence at higher levels is also harder, costlier, and slower to obtain. Evaluators must therefore choose, deliberately, how high they want to climb.

The seven steps of Bennett’s Hierarchy 🔗

The hierarchy moves sequentially through seven levels, as documented in the Journal of Extension:

1. Inputs: Resources invested in the program – funds, personnel, time, equipment, and knowledge base.

2. Activities: The educational events, workshops, demonstrations, and counselling sessions the program conducts.

3. Participation: The number and characteristics of people reached, plus the frequency and intensity of contact.

4. Reactions: Participants’ satisfaction with the program, their interest level, and their feelings about the activities and educational methods.

5. KASA – Knowledge, Attitudes, Skills, and Aspirations: Changes in what participants know, how they feel, what they can do, and what they aspire to. This is the most distinctive level of Bennett’s model and the bridge between process and outcome.

6. Practice Change: Actual behaviour change – what participants do differently after the program. A mother who not only knows about exclusive breastfeeding (KASA) but actually practises it falls here.

7. End Results (SEEC – Social, Economic, and Environmental Conditions): The ultimate, long-term changes in community-level conditions. Reduced infant mortality, improved nutritional status, lower fertility rates, or higher household incomes belong at this top step.

Process versus outcome evaluation 🔗

A useful feature of Bennett’s Hierarchy is that it cleanly splits process from outcome. The first four levels (inputs through reactions) measure how a program is being implemented – process evaluation. The last three levels (KASA, practice change, and SEEC) measure whether the program is producing the changes it promised – outcome evaluation.

This distinction matters enormously for population and family health programs. A POSHAN Abhiyaan district might run thousands of growth-monitoring sessions (strong process indicators) yet show little improvement in stunting (weak outcome evidence). Bennett’s framework reveals precisely where the chain is breaking – perhaps KASA is changing but practice is not, signalling barriers like poor food availability or gender-based decision-making norms.

The cost-evidence trade-off 🔗

Bennett was a realist about the cost of climbing the hierarchy. Measuring inputs and participation is cheap – registers and attendance sheets do the job. Measuring KASA needs pre- and post-tests. Practice change requires follow-up after several months. SEEC outcomes demand longitudinal studies, comparison groups, and sometimes sophisticated statistical techniques like analysis of covariance. Each step up multiplies the time, skill, and money required.

This is why Bennett urged evaluators to ask, before they begin: how high in the hierarchy do we realistically need to go, and how high can our budget take us? Decision-makers should be honest about whether they can accept softer evidence at a lower level – or whether they must secure resources to climb higher.

Comparing the two approaches 🔗

The LFA and Bennett’s Hierarchy are often presented as alternatives, but they are better understood as complementary. LFA is stronger at program design and management, especially for multi-sector projects with diverse activities. Its matrix forces clarity on assumptions and risks. Bennett’s Hierarchy is stronger at evaluating educational and behaviour-change programs, particularly in extension, health education, and nutrition contexts where KASA-to-practice transitions matter most.

In practice, many evaluators borrow from both. They use a logframe to plan and manage the program, while Bennett’s seven steps guide their evaluation design for measuring participant-level change. The Targeting Outcomes of Programs (TOP) model developed by Bennett and Rockwell explicitly merges this thinking by mirroring program planning and program performance.

Challenges in real-world evaluation 🔗

No matter which framework an evaluator chooses, certain challenges show up repeatedly in population and family health work. Recognising them in advance is half the battle.

Establishing valid control or comparison groups 🔗

The cleanest way to know whether a program caused an outcome is to compare participants with a similar group that did not receive the intervention. In practice this is hard. As field experience from MEASURE Evaluation documents, randomly assigning families to receive or not receive a health program is often ethically and politically impossible. Programs roll out across whole districts or states, leaving no obvious untreated comparison area. Even when comparison groups can be identified, contamination – where information or services spill over from program areas to control areas – can blur the difference.

Evaluators address this with quasi-experimental designs like propensity score matching, difference-in-differences, and stepped-wedge rollouts, but each comes with its own assumptions and limitations.

Measuring long-term impact 🔗

End results – reduced mortality, improved nutritional status, sustained fertility decline – take years to materialise. Yet funders, governments, and academics often want answers within a project cycle of three to five years. Researchers evaluating large-scale public health policies have flagged this tension between long-term goals and short-term interventions as one of the central difficulties of impact evaluation.

Additionally, the costs of follow-up data collection rise over time, while interest in old programs fades as new ones take their place. Many long-term evaluations are simply never done, leaving entire policies on permanent guesswork.

Securing accurate baseline data 🔗

You cannot prove change without a starting point. Yet evaluators frequently find that interest in formal evaluation arises years after a program has begun, eliminating any possibility of clean baseline data. Even when baselines are collected, timing matters: a baseline measured after sensitisation activities have already started will understate the program’s true effect.

Attribution and confounding factors 🔗

Population health outcomes are shaped by dozens of forces beyond any single program – economic growth, education levels, women’s empowerment, urbanisation, climate, and parallel government schemes. Public health evaluators routinely struggle to disentangle a program’s contribution from these moderators. A drop in infant mortality during a nutrition program’s tenure may owe as much to rising household incomes as to the program itself.

Heterogeneity of impact 🔗

Average effects can hide important variation. A family planning program may work well for educated urban women but poorly for adolescents in tribal districts. Evaluations that report only averages risk missing where a program is succeeding, where it is failing, and where redesign is needed most.

Choosing the right approach 🔗

For students and practitioners in population and family health studies, the lesson is not that one framework is better than another. It is that evaluation design must follow evaluation questions. If the question is “are we managing this program well?”, a logframe will serve. If the question is “did this educational intervention change what mothers know, feel, and do?”, Bennett’s Hierarchy is sharper. If the question is “did this policy reduce maternal mortality?”, neither framework alone is enough – robust counterfactual designs and longitudinal data are also needed.

Above all, good evaluation begins at the planning stage, not after the program ends. Building indicators, baseline measures, and comparison strategies into the program from day one is the single most important step toward credible answers later.

What do you think? If you were evaluating a national adolescent reproductive health program in your district, would you climb all the way to Bennett’s seventh level (SEEC outcomes), or stop at KASA and practice change? And how would you defend that choice to a funder demanding proof of long-term impact within three years?

References

1. https://www.betterevaluation.org/methods-approaches/methods/logframe
2. https://wikis.ec.europa.eu/spaces/ExactExternalWiki/pages/50108980/Logical+Framework+Approach
3. https://www.sopact.com/use-case/logframe
4. https://www.enablersofchange.com.au/bennetts-hierarchy-a-simple-yet-powerful-evaluation-framework/
5. https://archives.joe.org/joe/2010december/tt1.php
6. https://edis.ifas.ufl.edu/publication/WC092
7. https://www.data4impactproject.org/wp-content/uploads/2021/02/wp-14-157-1.pdf
8. https://pmc.ncbi.nlm.nih.gov/articles/PMC1937000/
9. https://www.data4impactproject.org/publications/impact-evaluations-of-large-scale-public-health-interventions-experiences-from-the-field/
10. https://publichealth.tulane.edu/blog/public-health-program-evaluation/