Regression Analysis Service: Linear, Logistic, Moderation and Mediation Models
Model how outcomes change with your predictors, and understand what the coefficients really say. We build, test and interpret regression models for research papers, theses and programme evaluations.
Regression models we build
- Simple and multiple linear regression, including hierarchical (blockwise) entry
- Binary, ordinal and multinomial logistic regression with odds ratios
- Poisson and negative binomial models for count outcomes
- Moderation (interaction) and mediation analysis, including bootstrapped indirect effects
- Dummy coding of categorical predictors and handling of control variables
- Multilevel and panel models on request
Diagnostics: making sure the model can be trusted
A regression that runs is not automatically a regression that is right. We check linearity, the distribution and variance of residuals, multicollinearity (variance inflation factors), influential cases and, for logistic models, fit and classification accuracy. When a diagnostic fails we fix the model, for example by transforming a variable, using robust standard errors or removing a predictor, and we document what changed.
Interpreting coefficients, R-squared and odds ratios
We explain every number in plain language: what a one-unit change in a predictor does to the outcome, how much variance the model explains, which predictors matter most and how confident the estimates are. For logistic models we translate odds ratios into statements a non-statistician can use. Before you collect data, the regression sample size guide helps you plan how many respondents you need for a given number of predictors.
Moderation and mediation
Many research questions ask not only whether X affects Y but why and for whom. Mediation tests whether an intermediate variable carries the effect; moderation tests whether the effect differs across groups or levels of another variable. We set up, test and chart these models, and report indirect effects with confidence intervals. For models with several latent constructs, see structural equation modelling.
For testing simple group differences first, see hypothesis testing.
How the work runs
- You share the dataset and describe the outcome, predictors and any theory behind them.
- We agree the model specification and check the data for problems.
- We fit the models, run the diagnostics and compare alternatives.
- You receive coefficient tables, charts, interpretation and the code or syntax used.
Frequently asked questions
How many predictors can I include?
It depends on sample size and the type of model. A common guide is at least 10 to 20 observations per predictor for linear models, and at least 10 events per predictor for logistic models.
What is the difference between linear and logistic regression?
Linear regression predicts a numeric outcome such as a score. Logistic regression predicts the probability of a category, such as pass or fail, using odds ratios.
My R-squared is low. Is the model useless?
Not necessarily. In social science data a low R-squared is common. What matters is whether the predictors are significant, the effects are meaningful and the diagnostics are sound.
Can you test mediation and moderation?
Yes, including bootstrapped indirect effects and interaction plots.
Do you also run regression in STATA and R?
Yes. We use SPSS, STATA, R or Python and provide the syntax or script.
Related statistical analysis services
Hypothesis Testing
t-tests, ANOVA, chi-square and non-parametric tests with assumption checks.
Structural Equation Modelling (SEM & PLS-SEM)
CB-SEM and PLS-SEM for complex theories with several constructs.
Factor Analysis (EFA & CFA)
EFA and CFA to find and confirm the structure behind your items.
Statistical Analysis Services
See every statistical method we support.