Released Exploratory Desktop v16.2🔥

News
#1

We have released Exploratory Desktop v16.2 on September 12th, 2026! :fire:

This release includes enhancements such as support for Latent Class Analysis (LCA), which can identify hidden segments within categorical data, the ability to create weight columns and perform weight-based aggregation in Pivot Table, and support for “Repeat By” (grouping) in Number charts!

Please upgrade to the latest version and give it a try!

Please see below for the new features and bug fixes included in this release.

Product Enhancement

Data Wrangling

  • Filter: We added support for “Any of the columns” and “All of the columns” operations in the Filter step. Users can now easily create complex filters, such as removing rows where any of the selected columns contain an NA, with a single operation.
  • We introduced a “Create Weight Column” feature that allows users to generate weights based on population composition. This enables users to perform weight-back adjustments easily, ensuring their sample analysis reflects the target population.
  • We introduced a new UI to unite multiple 0/1 dummy columns into a single comma-separated column. Users can now easily transform survey data into a readable format.
  • We redesigned the step node tokens in the Step Diagram for better readability and removed unnecessary lines when selected.

Chart

  • Scatter: We implemented a comprehensive Jitter feature for Scatter plots that is deterministic and visualization-only. Users can now clearly see the distribution of overlapping points in discrete data without affecting the underlying statistical analysis.
  • Scatter: We enhanced the LOESS trend line to support family, degree, and span arguments. Users can now fine-tune the smoothness and robustness of trend lines directly from the UI, allowing for more precise modeling of non-linear relationships.
  • Number: We added “Repeat By” support for Number charts in dashboards and notes. Users can now automatically generate multiple number tiles based on a grouping variable, making it easier to compare key metrics across different segments.
  • Number: We added a “Visualization” setting to the Number chart’s Value shelf to support indicator bars for ratios. Users can now visually represent proportions directly within the chart, making it easier to compare values at a glance.
  • Number: We added an option to show the row count as a sub-metric and a ratio indicator for Number charts when the main metric is a Ratio. This provides users with immediate context regarding the sample size and composition behind the ratio.
  • Number: We enabled the creation of multiple Number charts for different columns at once. Users can now select multiple metrics to generate a series of number tiles simultaneously, significantly saving time when building dashboards.
  • Number: We updated the thumbnail capturing framework so that it can now show sub-metrics in the thumbnails.
  • Pivot Table: We added drag-and-drop support for switching column assignments between Row, Column, Value, and Repeat By shelves. This allows users to experiment with different chart layouts much faster and more intuitively.
  • Pivot Table: We added support for weight-based aggregation in Pivot Tables. Users can now incorporate population weights into their summaries, enabling more accurate and representative analysis of survey data.
  • Pivot Table: We added an option to remove NA values directly within the “Limit Values” setting. Users can now filter out missing data more easily, especially when working with “Variable Values” in complex tables.
  • Pivot Table: We added an option to remove specific values after using the “Multiple Values” (separate values) option in Limit Values. Users can now exclude irrelevant categories from multi-answer survey data more easily.
  • Radar Chart: We added text wrap support for axis labels in Radar charts. This allows users to display long category names clearly without them overlapping or being cut off, improving the readability of the chart.
  • Repeat By: We added the ability to export images for each individual chart generated by the “Repeat By” function. Users can now easily save and share specific segment charts for use in external reports and presentations.
  • Chart: We implemented a smart auto-fit mechanism for axis labels when “Wrap Long Labels” is enabled. The system now automatically adjusts font size and angles, ensuring that charts remain clean and professional regardless of label length.
  • Heatmap: We updated the Heatmap chart to support displaying both the count (N) and the ratio simultaneously within the cells.
  • AI Summary: We enabled background content scrolling while the AI Summary dialog is open. Users can now refer to the underlying charts while reading summaries.
  • We added a “Paste” option to the chart menu. Users can now duplicate charts using a familiar menu interface in addition to the existing drag-and-drop method.

Analytics

  • Latent Class Analysis (LCA): We implemented support for Latent Class Analysis (LCA). Users can now perform clustering using categorical data, allowing them to identify hidden segments in survey responses and other non-numeric datasets.
  • ANCOVA: We redesigned ANCOVA’s analytics report.
  • ANCOVA: We overhauled the ANCOVA R layer to support multiple covariates and implement a more robust model selection logic. Users can now perform more complex and statistically sound covariance analyses with improved accuracy and reliability.
  • ANCOVA: We added support for 95% Q-Q Envelopes in ANCOVA using parametric bootstrapping. Users can now visually assess whether residuals deviate significantly from a normal distribution, providing a more reliable basis for model validation.
  • CHAID: We added support for numerical target variables in CHAID decision trees using ANOVA F-tests. Users can now identify key drivers for continuous metrics like revenue or scores, enabling more flexible segment analysis.
  • K-Means Clustering: We added a chart to show which variables are more significant or influential for creating the clusters. Users can now easily understand which variables differentiate their clusters, making the clustering results more interpretable and actionable.
  • K-Medoids Clustering: We improved the Radar charts in K-Medoids clustering to better group and visualize data. Users can now more easily identify and compare the characteristics of different clusters.
  • Decision Tree: We updated the decision tree visualization to show operators on the left for “Less” and on the right for “Greater,” and fixed the sorting of numeric split intervals. Users can now read split conditions more intuitively and see values in a logical ascending order.
  • Cronbach’s Alpha: We updated Cronbach’s Alpha to retain all values for Factor type variables. This ensures that users working with categorical data maintain full data integrity during reliability analysis.
  • Statistical Test: We redesigned the statistical test menu by grouping tests into logical categories and adding descriptive tooltips. Users can now more easily find and understand the appropriate test for their analysis.
  • Chi-Square Test: We updated the Chi-Square test report to display N and Ratio as columns rather than rows.

Issue fix

AI Data

  • Common R functions like read_csv() were not being correctly syntax-highlighted in the R Script editor within the AI Data dialog.
  • The Snowflake ODBC driver auto-setup failed because the vendor (Snowflake) changed the package naming convention and driver internals.

Data Source

  • Google BigQuery queries using session parameters returned no data after import, despite showing correct results in the preview.

Data Wrangling

  • The “Month - Name (1-12)” date filter failed to function on macOS while working correctly on Windows.
  • The “Cached at” timestamp failed to update when a user re-ran a step that had a stale (yellow) cache.
  • The “Replace Values” dialog failed to show other Factor columns for selection even when the target column was a Factor type.
  • The search dropdown in the Filter dialog closed prematurely when a Japanese user confirmed an IME character conversion.
  • Tooltips in the Step Diagram were not displayed when users hovered their mouse over the header area of certain nodes.

Chart

  • Charts in “Repeat By” layouts overlapped significantly when the number of charts per row was set to 1 or 2, particularly with many facet values.
  • Scatter charts ignored color opacity settings after a user clicked a data point to open the “Show Detail” dialog.
  • Switching between chart tabs was slow, likely due to a regression in the thumbnail capturing process.
  • The Pivot Table column total failed to update correctly when a Window Calculation was used in combination with “Limit Values.”
  • The Pivot Table ignored Factor level order for “Variable Values” when multiple variables were selected, resulting in an incorrect display order.
  • The thumbnail images for Number charts were displayed too large in the interface, and borders were missing from Pivot and Table thumbnails.

Analytics

  • Decision Tree: In CART decision trees, the “TRUE” nodes were not consistently placed on the left side after the first split.
  • Decision Tree: The CHAID tree diagram had incorrect category ordering for age groups and was missing values in the tooltips.
  • ANCOVA reports showed inconsistent row counts between unadjusted and adjusted statistics when covariates contained NA values.
  • ANOVA models failed with opaque errors when the category column was a logical, Date, or POSIXct type.
  • Correspondence Analysis models with only one dimension displayed a raw R variable name instead of the standard “Dimension 1” label in the coordinate table.
  • Logistic and Ordinal Logistic Regression models incorrectly attempted to display multicollinearity information when only one variable was present, resulting in an error.
  • One-way ANOVA failed with a raw error when at least one group was a singleton, even if other groups had sufficient data.
  • Factor Analysis: Selecting the “Geomin (Oblique)” rotation in Factor Analysis caused the analysis to crash due to an incorrect internal parameter name.
  • Several two-group statistical tests crashed with opaque errors when data filtering removed all observations for one group in ungrouped analyses.
  • Statistical tests failed to validate column types at runtime, allowing non-numeric data to produce invalid results after upstream configuration changes.
  • Statistical values on probability distribution charts for t-tests, ANOVA, and Chi-Square tests were incorrect and showed random vertical lines.
  • The “Max Iteration Times” setting in K-Means clustering was ignored, causing the analysis to always run with a default number of iterations regardless of user input.
  • The “Pair List” table in Chi-Square tests was displaying counts as decimals instead of integers and needed rounding for other statistics.
  • The “Reverse-code items automatically” option in Cronbach’s Alpha failed to affect Standardized Alpha when using Polychoric or Mixed correlation methods.
  • The aggregated t-test displayed the configured null mean difference but ignored it during calculations, resulting in statistics and p-values based on a zero difference.
  • The aggregated two-sample t-test failed to handle cases where both groups had zero variance, leading to raw internal errors or invalid statistics.
  • The ANCOVA “Means (Adjusted)” chart appeared empty when no covariate was selected due to a column naming mismatch in the R preprocessor.
  • The ANOVA test returned raw R errors instead of user-friendly messages when covariates were constant or collinear and interaction was disabled.
  • The application experienced significant performance delays when opening the Analytics view or deleting an existing Analytics item.
  • The CHAID decision tree report incorrectly converted commas in column names into periods, causing the names to be displayed incorrectly in tables.
  • Correlation: The correlation analysis summary was displaying incorrect information due to a mismatch between analysis conditions and the data output.
  • The K-Means Elbow method crashed on datasets with many duplicate rows because the candidate cluster counts exceeded the number of distinct observations.
  • The Kaiser criterion column in the PCA eigenvalue table appeared blank when the “Normalize” option was turned off.
  • The Kruskal-Wallis test failed during post-hoc testing when the category column was not a factor type (e.g., numeric or date).
  • The Normality Test aborted for all selected columns if even one column encountered an error, such as having too few observations.
  • The Normality Test failed to provide a result or crashed when a selected column was degenerate (contained only one distinct value).
  • The Normality Test performed sampling before filtering out NA values, leading to inconsistent results and potential failures depending on the random seed.
  • The Normality Test UI allowed sample sizes below the minimum requirement of 3, leading to runtime errors.
  • The one-sample proportion test encountered errors when the benchmark probability was 0 or 1 because the underlying R function returned logical instead of numeric p-values.
  • The one-sample t-test incorrectly propagated infinite values into summary statistics instead of excluding them, leading to invalid results.
  • The PCA Signed Loadings chart failed with an error when analyzing categorical variables using the Pearson correlation method.
  • The Radar chart in K-Means Clustering failed to render correctly for certain datasets.
  • The residual lines in the ANCOVA “Residuals vs. Fitted Values” chart were being displayed incorrectly.
  • The submenus for “Power Analysis” and “A/B Tests” were being cut off at the bottom of the screen in the Analytics Type dropdown.
  • Variables in the K-Medoids Radar chart were not sorted by similarity, making the chart harder to interpret compared to K-Means.

Dashboard

  • Clicking on dashboard tabs caused a blue focus ring to appear and persist, creating a distracting visual artifact.

Parameter

  • Dashboard parameter value lists were reverting to obsolete states during publishing or upon initial access on the server.

Project

  • Closing certain projects took an excessive amount of time (e.g., 25 seconds) due to performance inefficiencies.