So empirical researchers understand the analysis process

Statistical analysis in SKARI does not jump straight to a results table once a method is chosen; the whole analysis is organized as a six-step flow. Each step explains which variables to choose, which settings to use, and how those settings can affect the result. So instead of simply pressing a button to run an analysis, users can proceed while understanding what they are setting right now and why.

When the analysis finishes, instead of showing complex statistical tables first, results are presented in this order: a summary of the results, an interpretation tied to the research purpose, explanations of key indicators, and then detailed tables and charts. It first briefly shows what this analysis confirmed, then explains how that result relates to the research purpose or hypothesis you set out to test. After that, you can move into the detailed results as needed to check statistics, test results, charts and tables.

The interpretation is not the AI reading back the numbers shown on screen. It is designed so that the analysis and test results actually computed on the backend are structured and passed along, and the meaning of the results is explained on that basis.

Each table, too, goes beyond presenting numbers: it explains what the table is meant to check, which columns and statistics to look at, what the values mean, and how they can be interpreted in the research findings.

Recently, beyond standard 2D charts, we have been adding 3D visualizations for certain analyses, along with simulation and prediction-profiler features that let you change variable values directly and see how the results change.

And every analysis comes with the R or Python code actually used in the computation. The variables and settings chosen on screen carry over into the code as-is, and the code is annotated to match the six-step flow, so you can follow how each setting turns into a function and option. Because results are never left as a black box, when an advisor or committee member asks how a number was obtained, you can show the analysis process exactly.

Running this code produces the same result as the screen. R and Python are free for anyone to use, so results can be reproduced without any particular program, and the code can go straight into a paper’s appendix or answer data-and-code sharing requirements. Researchers also pick up the code naturally as they work, which helps when they later move on to more complex analyses.

# 1. Variables: group and value (weight)
data <- PlantGrowth
data$group <- factor(data$group)

# 2. Settings: standard ANOVA + Welch, effect sizes η²·ω²
library(car)
library(effectsize)

# 3. Data check: equal variances (Brown-Forsythe) and residual normality
leveneTest(weight ~ group, data = data, center = median)
fit <- aov(weight ~ group, data = data)
shapiro.test(residuals(fit))

# 4. Summary: ANOVA and Welch test
summary(fit)
oneway.test(weight ~ group, data = data, var.equal = FALSE)

# 5·6. Effect sizes and post-hoc test
eta_squared(fit)
omega_squared(fit)
TukeyHSD(fit)

            Df Sum Sq Mean Sq F value Pr(>F)
group        2  3.766  1.8832   4.846 0.0159 *
Residuals   27 10.492  0.3886

Connecting empirical survey research with statistical analysis

Empirical research using surveys is extremely common across fields. Yet on typical survey platforms, after collecting responses you often have to download the data as Excel or CSV and then move to another program such as SPSS, R or Python to run the statistical analysis.

df <- read.csv("responses.csv")
df$Q7 <- recode(df$Q7,
  "Strongly agree" = 5, ...)
df$Q2 <- factor(df$Q2)
# Decide yourself which analysis fits
summary(aov(Q7 ~ Q2, data = df))

SKARI aims to connect survey design, data collection and statistical analysis into a single research process. Once a survey is complete, you can check the data structure and response distributions right in Survey Lab without moving the responses to another program, and SKARI also recommends which statistical analyses can be applied based on the collected data and variable types. For example, you can see — according to the data structure and research purpose — whether you need to check differences between groups, analyze relationships between variables, or run regression or factor analysis.

Beyond standard multiple-choice and Likert-scale surveys, SKARI also supports specialized survey techniques for different research purposes, such as AHP, Conjoint, MaxDiff, Kano, NPS and price sensitivity analysis. Methods that would otherwise need separate specialist software or complex design are packaged as templates, connecting design, response collection and analysis.

When a survey closes, the results screen first summarizes things like response count, completion rate, margin of error and average response time automatically. For each item, the response distribution comes with an interpretation of how to read it, and you can cross two items to compare right away how responses differ.

After a survey closes, SKARI provides a separate Research Mode in addition to the basic results screen. In Research Mode you can run the statistical analyses your research needs on the spot, without downloading the response data. Everything the full survey research process requires is ready — from descriptive statistics for understanding response distributions, to scale reliability and factor structure, tests of group differences, regression and driver analysis to find the factors behind outcomes, and response quality and missing-response handling. Each result comes with an explanation of what it shows and how to read it, so it leads straight into analysis you can use in a paper, beyond a simple response summary.

Understanding response distributions

Descriptive statistics, radar chart, crosstabs (chi-square test), association rules (market basket analysis)

Reliability and scale structure

Reliability (Cronbach α, reliability by dimension), factor analysis, correlation matrix

Differences between groups

t-test, Welch test, Mann-Whitney test, analysis of variance (ANOVA), Kruskal-Wallis test, Tukey post-hoc test

Factors driving outcomes

Multiple regression, driver analysis

Change over time

Trends by segment

Data quality

Response quality diagnostics, response time analysis, missing-response handling

Machine learning analysis and interpretation for empirical researchers

Social science research increasingly uses machine learning for prediction, classification and variable importance alongside traditional statistics. But existing AutoML tools often focus on training models automatically and comparing performance, so researchers unfamiliar with machine learning can find it hard to understand why a certain variable matters or why a certain observation was predicted the way it was.

SKARI Auto Modeling trains and compares multiple machine learning models automatically, and focuses on explaining the model’s decisions and predictions so researchers can understand them. Depending on the classification or prediction goal, you can train models such as Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, SVM, KNN and XGBoost, and compare their performance with cross-validation and evaluation metrics. The whole process runs in order: data check, feature setup, model training, performance comparison, model explanation, prediction and simulation.

Once a model is chosen, you can understand the results step by step through a range of explanation techniques. Each screen shows the conclusion before the chart, and the chart follows as evidence for that conclusion. SHAP importance shows which variables had the biggest influence on the model’s predictions across the whole dataset and in which direction, while PDP and ALE show in which direction the model’s predictions change overall as a particular variable changes.

The SHAP waterfall explains, variable by variable, why a particular individual or observation received its prediction, and What-if Simulation shows how the prediction changes when you change the value of just one variable while keeping that observation’s other conditions fixed. With SHAP Interaction, you can explore how combinations of two variables affect the model’s predictions.

Finally, it tells you what to check before moving results into a report. It keeps you from reading predictive relationships as causal, and points out that the effects of related variables can appear to overlap and that explanations were computed on a sample rather than the whole dataset. Then, using only the preceding results and no newly computed numbers, the AI summarizes everything in one paragraph that researchers can use as a first draft of their results write-up.

From the theoretical model of empirical research to surveys and structural equation analysis

Structural equation modeling (SEM) is widely used in business, education, psychology and the social sciences to test theoretical relationships. But in real research, running an SEM is not the end of the study. Researchers first review prior work and theory, define constructs and the relationships between variables, form research hypotheses, and then write survey items to measure them. After that, they collect data, verify reliability and the measurement model, and then analyze the structural model.

SKARI SEM Lab connects this process into a single research flow. When a researcher enters a research topic or question, AI helps review related theories, constructs, relationships between variables and research hypotheses, and draft the theoretical model.

Based on the designed research model, you can write survey items to measure each construct and run the actual survey through Survey Lab. When the survey is complete, you can connect the research model and survey structure to SEM Lab for analysis, without downloading the collected data again or setting up variables from scratch.

download the CSVassign the variables againre-link the constructs— none of these three

When the analysis is done, the results are organized into a report of nine chapters. It runs in the order of sample and data, descriptive statistics, reliability, measurement model, model fit, structural model, effect decomposition, diagnostics and limits, and AI interpretation — structured to follow how the results chapter of a paper is written, so it can serve directly as the skeleton of your results write-up. Each chapter shows what it checks and the criteria for judging it, and if any result falls short of a criterion, it tells you first, at the top of the chapter. Below is an example of the measurement model chapter.

Another important feature is that you do not have to write lavaan syntax yourself. When a researcher builds the research model on screen by connecting latent and observed variables and drawing paths between them, SKARI converts that structure into an analyzable lavaan model. You can design and analyze a research model visually without writing code, and after the analysis the R code actually used is provided as well, so the results can be reproduced.

Besides drawing directly on screen, you can also upload an image of a research model drawn by hand on paper or a tablet. SKARI recognizes the latent variables, observed variables and paths in the image and turns them into an editable research model and a lavaan model, so researchers can sketch the model the way they are used to and start the analysis right away.

→

Recognized model

3 latent variables · 9 observed variables · 3 paths

# Measurement model
VisualAbility  =~ x1 + x2 + x3
TextualAbility =~ x4 + x5 + x6
SpeedAbility   =~ x7 + x8 + x9
# Structural model
TextualAbility ~ VisualAbility
SpeedAbility   ~ TextualAbility + VisualAbility

Edit on canvas