How to measure whether digitalisation reduces or widens social divides

Talking about the digital divide as though it were a single difference between people who are connected and those who are not oversimplifies the problem. Access matters, but so do the devices people have, whether their connection is suitable for the tasks they need to perform, the skills they possess, and the activities they are able to carry out. Someone may have an internet connection and still struggle to complete an administrative procedure, learn online, or protect their data. Another person may use the internet frequently but lack the tools or support needed for a particular task.

That is why it is useful to distinguish between dimensions before comparing groups or evaluating a policy. Connectivity describes a condition of access, not the social outcome of that access. Nor does frequency of use, by itself, show that an activity is beneficial, safe, or independent. These questions are related, but they are not equivalent; combining them in a single indicator can conceal which barrier remains and whom it affects.

A useful analysis distinguishes at least between the availability of a connection and devices, the skills needed to navigate digital environments, types of use, and outcomes. The last dimension includes questions such as whether someone can access a service, meet a need, or take part in an activity. To claim that inclusion exists, it is not enough to confirm that technology is present: it is necessary to examine whether it enables people to do what it is intended to help them do.

This distinction helps avoid two common mistakes. The first is interpreting better access as automatic proof that gaps have closed. The second is attributing inequality to a lack of skills when the data record only availability or frequency. Every conclusion must be limited to what the indicator actually measures and to the population included in the measurement.

In Spain, a recent reference source is the National Statistics Institute’s (INE) Survey on Equipment and Use of Information and Communication Technologies in Households, whose results note relates to 2025. The availability of a recent edition makes it possible to follow the issue using official information, but the survey’s title should not be mistaken for a complete description of digital inclusion: the tables, definitions, and methodology for each specific variable must first be consulted.

When analysing the results, the initial question should be precise: is the measurement about the presence of equipment, internet access, an activity carried out, or a particular skill? These are not interchangeable indicators. Before comparing two percentages, it is also necessary to confirm that they refer to the same period, population, age group, and definition. If categories or the questionnaire change, a difference between editions may partly reflect a change in measurement rather than a social transformation alone.

The European Union’s Digital Decade provides another framework: Decision (EU) 2022/2481 establishes the European strategic programme for 2030. That horizon helps place objectives and monitoring within a broader public policy, but it does not replace consultation of national indicators, nor does it turn every statistic about household use into a direct measure of progress towards the programme’s targets.

A reading sheet should record at least the indicator, its definition, the reference period, the population to which it applies, and the available breakdowns. It should also clarify whether the figure comes from a survey response or another type of observation. A number makes sense alongside its denominator and method, not as an isolated figure repeated in a headline. If a result is not broken down by a particular characteristic, it is not possible to infer how it is distributed within that group.

Comparisons by age, income, location, or other characteristics can show where differences in a variable are concentrated. But observing that two groups have different results does not establish why that is the case. Several factors may be operating at the same time—such as education, occupation, device availability, or support needs—and a descriptive survey may not be able to distinguish their respective influence. The data can help formulate questions and locate observed inequalities; they cannot assign causes without further analysis.

Before claiming that a gap affects a group, it is necessary to check that the source publishes that breakdown and that the sample supports a meaningful interpretation. A broad category may conceal internal differences; an estimate based on few cases may be unstable. Age, income, and location are not self-sufficient explanations either. When several variables are compared at once, it is necessary to know whether the analysis can separate their associations or merely presents descriptive cross-tabulations.

It also matters which aspects are not observed. A general household survey does not automatically amount to a specific assessment of accessibility or the needs of people with disabilities. Likewise, recording that someone carried out an activity does not reveal whether they completed it independently, with another person’s help, or after several attempts. If the source does not ask about these details, a conclusion should acknowledge that they are missing rather than fill the gap with an intuitive interpretation.

A balanced reading distinguishes three levels: the difference shown by the data, the hypotheses it might suggest, and the evidence needed to test those hypotheses. An association between characteristics and use does not prove that one characteristic causes the other. This caution does not diminish the value of statistics; it prevents a useful description from becoming a causal claim unsupported by the measurement design.

A survey has a scope defined by its questionnaire, population, and reference period. Self-reported answers may not capture accurately the quality of a connection, the difficulty of a task, or the help received. In addition, published categories may group together different situations. These limitations do not invalidate the information: they explain what kinds of claims it can support and where it would need to be supplemented with other data.

Temporal comparisons require similar caution. A change between two years cannot be attributed to a specific public measure simply because both occurred at roughly the same time. To assess the effect of an intervention, data would be needed on who received it, when they received it, what changed, and how that change was compared with an appropriate reference situation. Without such a design, the result describes a development over time but does not, on its own, identify its cause.

It is also worth avoiding a single indicator of “digitalisation”. A percentage of households with access does not summarise skills; a measure of skills does not demonstrate that services are accessible; and use of a service does not guarantee that it produced the intended outcome. A policy may improve one dimension while leaving another unchanged. If a composite index is presented, its components, weightings, and construction choices should be made visible so that shortcomings are not hidden by an average.

Accordingly, figures should be read as elements of a diagnosis, not as an overall score of whether society is included or excluded. The absence of an observed difference does not prove that no difference exists: it may depend on how the question was worded, which groups were included, or what level of detail was published. Explaining these limits in an article is not a methodological weakness; it is part of the information that enables readers to judge the conclusion appropriately.

Research can help interpret what general statistics do not explain, provided its scale is respected. A 2023 study on digital inclusion in the education sector describes research focused on an educational community, using structured surveys and analysis alongside teachers. This design offers a situated case and combines quantitative and qualitative information; it does not, by itself, represent the whole Spanish population or allow its findings to be transferred directly to other settings.

This kind of evidence is valuable for raising questions: whether available access is maintained over time, which applications are used, or what difficulties arise during an educational transition. But observations from one community should not be presented as a national trend. The purpose of cross-checking is to complement official series and make possible mechanisms visible, not to replace a representative estimate with an example or generalise beyond the study’s design.

A responsible synthesis separates official evidence describing population patterns from research examining more specific experiences or settings. It also distinguishes sources by date: a 2021 report can offer historical context, but it is not enough to describe the current situation if later data are available. Results from different years, populations, or methods should not simply be added together; first, it must be explained what question each source answers and which comparisons are legitimate.

For readers, a practical check can be reduced to four questions: who was measured, what was asked, when was the information collected, and what inference does the design allow? If a publication does not answer those questions, it is premature to use it to support a broad claim. Comparing sources improves analysis when their differences are made explicit, not erased.

Claiming that a technology policy reduces gaps requires more than showing that general connectivity increased. It would be necessary to observe changes in the dimensions the intervention was intended to improve, identify the target population, and check whether outcomes are distributed equitably. If the objective is to make administrative procedures easier, for example, internet access may be a prerequisite, but the relevant indicator should also describe whether people can complete the procedure and what support they need.

Evaluation must be consistent with the intervention: explicit objectives, indicators linked to those objectives, and comparable measurement periods. When an analysis aims to attribute effects to a policy, it must explain how it distinguishes those effects from other concurrent changes. If only a descriptive before-and-after comparison is available, that should be stated clearly rather than presented as conclusive causal proof. Precision about the design matters just as much as the final percentage.

The most defensible conclusion is not that a single figure confirms inclusion or exclusion, but that statistics can identify dimensions and differences that need explanation. Spain has a recent official survey on household ICT equipment and use, as well as a European monitoring framework through 2030; turning those references into a social evaluation requires transparent definitions, relevant breakdowns, and outcome measures.

The useful question, then, is not only how many people are connected, but who can benefit from technology, for what purpose, and under what conditions. A policy reduces a gap when evidence shows relevant improvements for those who faced the barrier, and when that improvement is not inferred solely from an overall average. Access is an indispensable starting point; demonstrating inclusion also requires measuring capabilities, uses, and outcomes, stating limitations explicitly, and not claiming more from the data than they can show.