Every indexing system promises to connect a searcher with the right document at the right time. But how do we know whether it actually delivers on that promise? Asking “is this system good?” is too vague to be useful. A library committee, a database manager, and an end user will each define “good” differently. This is why evaluation in information science is structured into distinct levels, each answering a sharper question than the last. At the base, we ask whether a system works. Then we ask whether it works at a reasonable price. Finally, we ask whether it is worth having at all. These three levels-system effectiveness, cost effectiveness, and cost-benefit evaluation-form a logical ladder. Understanding where each level sits and what it measures helps you judge any indexing system, from a small institutional catalogue to a national bibliographic database.
Table of Contents
- Why levels of evaluation exist
- System effectiveness: does it retrieve well?
- Recall and precision
- Exhaustivity and specificity
- Beyond the two core ratios
- Cost effectiveness: does it retrieve well at a reasonable price?
- What counts as a cost
- A worked illustration
- Cost-benefit evaluation: is the system worth having at all?
- The challenge of valuing benefits
- Putting the three levels together
Why levels of evaluation exist
The framework of layered evaluation owes much to the work of F.W. Lancaster, whose writing in the 1970s shaped how libraries assess retrieval systems. Lancaster argued that an information retrieval system can be judged by considering three connected issues: how well it satisfies its objectives, how efficiently it satisfies them, and whether it justifies its existence at all. Each of those questions maps onto a deeper level of analysis. The structured approach to evaluation documented in Indian LIS coursework follows this same logic, treating performance and cost as the two basic factors any thorough study must address.
The reason these levels matter is practical. A system can be highly effective and still be a poor choice if it drains a budget. Another system might be cheap but so weak that users abandon it. Evaluating at only one level produces a distorted picture. By moving through all three, an evaluator builds a complete case for keeping, modifying, or replacing a system.
System effectiveness: does it retrieve well?
The first level asks the most fundamental question. Does the system retrieve what users need, and does it leave out what they do not need? This is the domain of performance measurement, and it is where the famous metrics of recall and precision live.
Recall and precision
Recall measures completeness. It is the proportion of all relevant documents in the collection that the system actually retrieves for a query. Precision measures purity. It is the proportion of retrieved documents that are genuinely relevant. A system that pulls every relevant record but buries it under hundreds of irrelevant hits has high recall and low precision. A system that returns only a handful of perfect matches while missing many others has high precision and low recall.
These two measures pull against each other. The landmark Cranfield tests led by Cyril Cleverdon established that there is an inevitable inverse relationship between recall and precision, and that systems typically operate at a recall ratio between 70% and 90% while precision sits much lower, often between 8% and 20%. The same project found that there is an optimum level of exhaustivity in indexing-indexing beyond a certain point does little to improve recall but seriously weakens precision.
Exhaustivity and specificity
Two indexing decisions directly shape these outcomes. Exhaustivity refers to how many concepts in a document the indexer chooses to represent. Specificity refers to how precisely the index terms describe those concepts. High exhaustivity tends to raise recall, while high specificity tends to raise precision. An evaluation at the effectiveness level often traces a system’s recall or precision failures back to choices made about these two variables.
Beyond the two core ratios
Effectiveness is not only about recall and precision. Lancaster proposed a broader set of criteria, including the coverage of the collection, the response time, the form in which output is presented, and the effort the user must invest. The user-oriented criteria for retrieval evaluation note that several of these are easy to measure-response time is directly observable, and user effort can be partly captured through the time spent formulating a query and examining results. Recall and precision remain the hardest to measure because they depend on knowing how many relevant documents exist in the first place.
This level also accommodates the user’s perspective. A search for “best smartphone” that returns thousands of loosely related results scores well on recall but frustrates a user who wanted a quick answer. Effectiveness, then, is not a single number but a balance tuned to what users actually want. A researcher preparing a state-of-the-art review needs high recall and will tolerate noise. A user checking a single fact prefers precision and dislikes clutter.
Cost effectiveness: does it retrieve well at a reasonable price?
A system can be effective and still be impractical. The second level introduces money into the equation. Cost-effectiveness analysis relates the performance of a system to the resources consumed in achieving it. It does not try to convert benefits into rupees. Instead, it compares the cost of competing options against a measurable unit of output, such as the cost per search or the cost per relevant document retrieved.
This distinction is important. Cost-effectiveness analysis is most useful precisely when analysts cannot place a monetary value on the benefits. It expresses cost in money but measures effectiveness in physical units, and because the two are incommensurable, the result is a ratio rather than a single combined figure. In information work, that ratio might be the average cost of running one search at a given monthly search volume.
What counts as a cost
Lancaster identified several components that an honest costing must include. These go well beyond the price of the software. They cover the cost of conducting a search, the user’s effort, the time needed to learn the system, the cost of retrieving the documents themselves, and the time a user spends from submitting a query to receiving an answer. Many of these are indirect and difficult to calculate, which is why economic assessment of retrieval systems is often more complex than measuring effectiveness.
A worked illustration
A concrete study makes this tangible. An evaluation of the PROBE retrieval program searching the ERIC database measured both performance and cost together. It found that descriptor-based searches achieved an average relevance ratio of about 55%, while natural-language searches reached only 42%. On the cost side, a descriptor search cost roughly $14.42 when 150 searches ran each month but dropped to $9.18 per search when 300 searches ran each month. The study concluded that the service was both satisfying users and cost-effective, because at sufficient volume it could become self-supporting at a modest fee.
Two lessons emerge. First, cost per search falls as usage rises, so the cost-effectiveness of a system depends heavily on how intensively it is used. Second, a more effective search method (descriptors) can also be a more economical one, which means effectiveness and cost are not always in conflict. There are, however, genuine trade-offs between input costs-the effort poured into careful indexing-and output costs-the effort users spend wading through poor results.
Cost-benefit evaluation: is the system worth having at all?
The final and most demanding level asks whether the system justifies its existence. Cost-benefit evaluation weighs the total cost of a system against the total value it delivers to its users and to the parent organisation. Where cost-effectiveness stops at performance per rupee, cost-benefit analysis tries to express benefits in the same monetary terms as costs so that the two can be directly compared.
Lancaster drew a clear line between the two in his foundational work. His analysis of cost-effectiveness and cost-benefit approaches distinguished between them and explored the relationship between costs, performance, and benefits, including the trade-offs between input and output costs. Cost-effectiveness looks inward at how efficiently the system performs. Cost-benefit looks outward at whether the system is a worthwhile investment compared to doing nothing or doing something else entirely.
The challenge of valuing benefits
The difficulty is obvious. How do you put a price on a researcher finding a crucial citation, a clinician retrieving the right treatment guideline, or a student locating a vital source? These benefits are real but resist monetisation. This is exactly why cost-effectiveness analysis is often used as a substitute or a first step. As broader economic evaluation literature explains, cost-effectiveness analysis frees the analyst from expressing all benefits in monetary terms and is frequently the opening stage of a fuller cost-benefit study.
The same problem appears across disciplines. In health economics, analysts often prefer cost-effectiveness ratios precisely because they sidestep the ethical and analytical difficulty of assigning a monetary value to outcomes such as a year of life. Indexing systems face an analogous problem: the value of well-organised knowledge is diffuse and long-term, spreading across every successful search a system enables over its lifetime.
Putting the three levels together
The three levels are not rivals but stages of one inquiry. Effectiveness tells you whether the engine runs. Cost-effectiveness tells you the fuel economy. Cost-benefit tells you whether the journey was worth taking. A complete evaluation moves from the technical core outward to the organisational and economic context. A system that is effective, affordable to operate, and demonstrably valuable to its community has cleared all three bars and earns its place. One that fails at any level signals where attention-and possibly redesign-is needed.
For anyone managing an indexing or retrieval system, the practical takeaway is to never settle for a single verdict. Measure performance with recall, precision, and the wider set of user criteria. Cost that performance honestly, including the hidden burden on users. Then step back and ask the hardest question of all: does the benefit to the people who use this system outweigh everything it costs to keep running?
What do you think? If you had to evaluate the search system at your own institution, which level would reveal the most uncomfortable truths-the effectiveness numbers, the real cost per search, or the question of whether its benefits truly justify the investment? And in an age of free web search, how should libraries make the cost-benefit case for maintaining specialised indexing systems at all?
References
- https://ebooks.inflibnet.ac.in/lisp7/chapter/evaluation-and-measurement-of-information-retrieval-system/
- https://www-nlpir.nist.gov/projects/irlib/pubs/cranv1p1/cranv1p1_text/01_002.txt
- https://sigir.org/files/museum/introduction_to_modern_information_retrieval/chapter_5.pdf
- https://www.betterevaluation.org/methods-approaches/methods/cost-effectiveness-analysis
- https://eric.ed.gov/?id=ED121261
- https://asistdl.onlinelibrary.wiley.com/doi/abs/10.1002/asi.4630220104
- https://www.sciencedirect.com/topics/economics-econometrics-and-finance/cost-effectiveness-analysis
- https://www.ncbi.nlm.nih.gov/books/NBK10253/

Leave a Reply