From Data to Evidence
Research does not end when fieldwork does. The analytical process begins there.
Data is first cleaned, verified, coded, weighted and prepared for analysis, then examined with the statistical methods the research question calls for.
Advanced analysis performed on faulty or incompletely processed data only produces a more complicated error.
Data Quality Before Analysis
The first thing we look at in a dataset is not the results. We check a set of things first, and we do not produce an analytical result before those checks are complete.
- Are there missing values?
- Are there logical contradictions?
- Are there duplicate records?
- Are the outliers real, or errors?
- Is the coding consistent?
- Have the weights been applied correctly?
A data story
The Same Rate, a Different Reality
Say unemployment is 20% in two regions. At first glance the result looks identical. A closer analysis shows otherwise: in one region unemployment is concentrated among young people, in the other it is higher among women. In one the problem may be an education mismatch; in the other it may be transport, care burden or employer demand.
The same percentage can hide different mechanisms. This is exactly where analytics earns its place: we ask not only "how much?" but "among whom?", "where?", "why?" and "under which conditions?".
Descriptive Analytics
We draw the picture accurately first. Frequencies, proportions, means, medians, distributions, cross-tabulations and subgroup comparisons establish the basic structure of the data.
Descriptive analysis is not our final stop. It is what makes the next set of questions answerable.
Inferential Statistics
To know whether a difference observed in a sample also exists in the population, we use inferential statistics: t-tests, ANOVA, chi-square tests, non-parametric tests, confidence intervals and hypothesis tests as required.
The aim is not to run a test. It is to understand whether the observed difference could have arisen by chance.
Regression & Explanatory Models
When we want to understand the relationship between a variable and an outcome, we use multivariate models.
An SME’s access to finance may not be explained by firm size alone. Sector, region, firm age, headcount, export status and whether the owner is a woman entrepreneur may all matter at the same time. Simple proportions fall short.
Our analytical capacity covers linear regression, logistic regression, multivariate modelling and interaction analysis as required.
A realistic example
Before Saying "Women-Led Firms Borrow Less"
Suppose a study finds that women business owners use credit at a lower rate. The first reading would be: women entrepreneurs have less access to finance.
Good analysis does not stop there. Are the firms smaller? Is the sector distribution different? Does firm age matter? Does collateral capacity vary? Does the regional distribution drive the result?
Adding those factors to the model may shrink the original difference, enlarge it, or change it entirely. Hence our basic principle: we do not accept the first observed relationship as the final result.
Segmentation
A total average sometimes hides the differences that matter most. We therefore examine the data by age, sex, region, income, education, firm size, sector and behavioural profile.
We do not stop at demographic segmentation. Where it helps, we build behavioural or analytical segments: high potential–low access, high satisfaction–low loyalty, high risk–low awareness. Profiles like these are often far more useful to a decision-maker.
Index & Indicator Development
Some studies cannot be measured with a single question. Institutional capacity, quality of life, satisfaction, social cohesion, digitalisation and sustainability all need several indicators assessed together.
Index design, indicator selection, standardisation, weighting and scoring approaches are used as needed. The aim is not to produce a score, but to build a measurement system a decision-maker can follow over time.
Reliability & Validity
It is not enough for a scale or indicator to produce a high score. The measurement has to be reliable and valid.
Cronbach’s Alpha, ICC, Kappa, EFA, CFA, SEM and ROC analysis are used as appropriate. These matter particularly in health research, social research, scale development and evaluation studies.
Advanced Analytics
Advanced analysis is used to answer the question properly, not to look complicated. Not every study needs an advanced model.
Where the research question requires it, our capacity covers logistic regression, multivariate analysis, factor analysis, SEM, survival analysis, mediation and moderation, ROC analysis and non-parametric methods.
Our principle is clear: use the most accurate model, not the most complex one.
Survival Analysis
In some studies time is the critical variable. How long before an athlete leaves the sport, an employee leaves the job, a business closes, or a patient reaches a follow-up endpoint.
Survival methods show us not only the final state but how the risk changes over time.
Mediation & Moderation
Suppose a training programme increases employment. The question may not end there. Did the effect run through a rise in confidence? That is a mediation question. Is the effect different for women than for men? That is moderation.
In social policy and programme evaluation especially, these analyses answer not only "is there an effect?" but "how, and for whom, does the effect appear?".
Data Visualization
Good analysis has to be told well. An analysis can be technically flawless, but if the decision-maker cannot follow the result its effect is limited.
Visualisation is therefore part of the analytical process: tables, charts, heat maps, dashboards, trend visuals and regional comparisons as required.
The aim is not to produce a graphic. It is to make the right message visible.
Dashboard & Decision Support
For some projects a one-off report is not enough. Where a field operation or an indicator system has to be monitored continuously, dynamic monitoring systems can be built with Power BI and similar tools.
The decision-maker can then follow regional differences, field progress, core indicators and target-versus-actual on one screen. This capability is described in full under Digital Research & Dashboards.
Analysis for Policy & Management
Analysis is done for a decision. For us a good analysis does not end with the sentence "this was found to be statistically significant". The next question is what it means.
In reporting we therefore connect the statistical result to policy, to operations, to strategy and to implementation. The real value of research is not making the correct calculation; it is supporting the correct decision.
