PurifyData Pro: Your First Screening Session

English neural AI narration (Kokoro-82M via kokoro-onnx 0.4.9, Heart (af_heart)); not a human recording.
Recorded from app v3.6.1 using only its built-in practice dataset.

01. Start without an account (00:00)

Welcome to Purify Data Pro. Let's walk through your first data screening session using the app's current interface and built-in example. You do not need an account or password. Statistical calculations run in your browser. Choose Open workspace for your own data, or Try sample data to practise. For help, open the three-dot menu. Under Learn, you will find Video walkthrough and PDF guide. We will practise first, then import a CSV file.

Steps:

1. Open the app in a current desktop browser, such as Edge or Chrome.
2. On Home, find Open workspace and Try sample data below the introductory information.
3. Use Open workspace for your own CSV. Use Try sample data for a safe practice session.
4. Keep your original data file separately; the app is not a permanent data-storage service.

Remember: The menu's Learn section contains Video walkthrough and PDF guide, with download options. This tutorial demonstrates interface operation; it does not prescribe deleting outliers, imputing missingness, or using identical cut-offs in every study.

02. Practise with the sample (00:30)

Click Try sample data. The workspace opens with sixteen example rows and seven columns. Five numeric variables are selected for screening. The sample contains missing values and an unusual score so that you can explore the outputs. The left panel controls the analysis, and the tabs across the workspace show different results. You can import your own CSV at any time.

Steps:

1. Click Try sample data on Home.
2. Check the Variables panel: the example has 16 rows, 7 columns and 5 detected numeric variables.
3. Use the left panel for settings and the workspace tabs for results.
4. The example is for learning; its results are not evidence about your own research data.

Remember: You can reopen the menu using the three-dot button in the top-right corner. Open workspace starts at the top without clearing an existing dataset.

03. Import and select variables (00:55)

For your own data, prepare a CSV with one header row, one case per row, and one variable per column. Click Import CSV in the Variables panel or in the menu. Choose the file from your device. In Select required variables, choose the numeric measures you intend to screen, then click Analyze selected variables. Do not treat participant identifiers or numeric category codes as continuous measures just because they contain numbers.

Steps:

1. Save a plain CSV with column names in the first row. Keep cases in rows and variables in columns.
2. Click Import CSV in the Variables panel, or use Import CSV in the menu.
3. Choose your file from your device. The demonstration reuses the built-in example as a CSV.
4. In Select required variables, select the numeric measures needed for the analysis.
5. Click Analyze selected variables. Inspect the resulting variable names and row count.

Remember: Exclude identifiers and variables inappropriate for continuous-data screening. If a measure is not detected as numeric, check for text, units, inconsistent decimal formatting, or non-numeric missing-value codes in your source file.

04. Review the screening settings (01:25)

After import, read the Screening Setting Notice. Then find the orange Data Screening Settings section in the left panel. Icons beside the titles help you recognise each setting. Missing-value handling decides whether incomplete cases are removed before screening. Z-scores and I Q R screen one variable at a time. Mahalanobis distance screens unusual combinations of selected numeric values. Review Delete flagged rows from screened exports separately. Open a guide beside a setting when you need help choosing its cut-off. A flag is a prompt for review, not proof of an error.

Steps:

1. Read the Screening Setting Notice after importing, then close it using its X button.
2. Find the orange Data Screening Settings section; its icon-labelled title bands identify the rules.
3. Review missing-value handling, Z-score and IQR cut-offs, and the export-deletion checkbox.
4. Scroll within the left panel to review Mahalanobis, skewness, kurtosis and normality case limits.
5. Expand a setting's guide for context, then choose settings that fit your research plan.

Remember: If missing-value deletion is off and selected values are missing, the Missing-value decision dialog offers Retain for manual review or Create imputed dataset. These are different choices; do not choose imputation merely to make a warning disappear.

05. Read the overview counts (02:04)

Open Overview. Rows is the number of cases in the current working dataset. Complete rows have values for every selected measure. Retained counts describe each workflow's screened export. Tested counts show the complete cases that received a Mahalanobis distance. These counts can differ when incomplete cases are retained. Expand Understanding overview counts for definitions. If a relevant detector is switched off, its cards indicate that the analysis was not applied rather than implying successful screening.

Steps:

1. Click Overview after choosing your variables and settings.
2. Read Rows, Variables and Selected numeric first.
3. Compare Complete rows with Multivar. tested and Combined tested.
4. Use the corresponding retained cards to understand the final screened counts.
5. Expand Understanding overview counts when a label is unfamiliar.

Remember: A case can be retained without being tested by Mahalanobis distance when missing-value deletion is off. Not applied does not mean zero outliers; it means the corresponding workflow was not applied. Read status wording as well as numbers.

06. Inspect one-variable outliers (02:39)

Click Univar. Outliers. This page screens each selected variable separately using the enabled z-score and I Q R rules. Review the summary table, the screening note, and the deleted and screened cases farther down the page. Check flagged values against the original records and your study context. If you change a rule in the left panel, the results update. Do not remove a valid observation solely because it is unusual.

Steps:

1. Click Univar. Outliers.
2. Inspect the screening table for the selected variables and enabled rules.
3. Read Univariate Screening Note to confirm the rules actually used.
4. Scroll to Univariate Deleted and Screened Cases for case-level review.
5. Use Export cases CSV when you need a separate case-review record.

Remember: Outlier screening and deleting flagged cases are not the same action. Review the export-deletion checkbox and substantively justify any exclusions. Keep a record of the original values and decisions.

07. Inspect score patterns and correlations (03:08)

Click Multivar. Outliers. Review Mahalanobis Distance and its tested-case information. It looks for unusual combinations across the selected numeric variables, rather than a single extreme value. Scroll to Correlation Matrix on this same page. High-correlation cautions identify pairs worth investigating; the correlation matrix does not itself delete cases. Incomplete cases may have no Mahalanobis value, so distinguish tested cases from retained cases when reporting the results.

Steps:

1. Click Multivar. Outliers.
2. Inspect Mahalanobis Distance, the selected cut-off and the number of tested cases.
3. Scroll to Correlation Matrix to review variable relationships and any high-correlation cautions.
4. Use Export matrix CSV if you need the matrix outside the app.
5. Read the screening note and case table before deciding whether exclusions are justified.

Remember: High correlation is a caution about relationships among variables, not an automatic outlier-deletion rule. No Mahalanobis value is different from a calculated non-outlier result.

08. Understand the combined workflow (03:43)

Combined Outliers is a separate workflow. It applies univariate screening first, then Mahalanobis screening to the remaining cases. Its counts and data basis can therefore differ from standalone multivariate screening. Open Combined Outliers and review its summary and screening note. Use the combined export only when this sequential workflow is the one you intend to report. Do not assume that all screened exports contain the same cases.

Steps:

1. Click Combined Outliers.
2. Read Combined Outlier Screening and its screening note.
3. Remember the order: Univar. Outliers first, then Multivar. Outliers on the remaining cases.
4. Compare its tested and retained counts with the standalone workflows.
5. Choose the export that matches your documented analysis plan.

Remember: Do not mix standalone Multivar. Outliers statistics with a Combined Outliers dataset without checking that they refer to the same cases and settings.

09. Check normality and the data basis (04:12)

Click Normality and wait for the calculation to finish. First check Normality Assessment Data Basis so you know which cases each assessment uses. Expand Decision rule to read the app's criteria. Scroll to the univariate results, histograms and Q Q plots, and the multivariate and combined assessments. Skewness and kurtosis describe distribution shape; they are not outlier-deletion rules. Read plots and supporting tests together. A normality label or one p-value does not settle every assumption of your planned analysis.

Steps:

1. Click Normality and wait until the progress indicator finishes.
2. Read Normality Assessment Data Basis before interpreting statistics.
3. Expand Decision rule to inspect the app's criteria and case-limit settings.
4. Scroll to the univariate table, histograms and Q-Q plots.
5. Review standalone multivariate and combined assessments separately.

Remember: Check the number of cases actually tested, especially when a case-limit mode is used. These are initial screening diagnostics, not a complete assessment of regression residuals or every assumption of a later statistical model.

10. Use recommendations as guidance (04:48)

Recommendations summarises the app's screening evidence. Start with Consolidated Decisions, then expand a relevant card under Recommended Statistical Analysis Path. The cards open independently, so you can focus on one topic at a time. Treat these recommendations as educational guidance. Your research question, measurement scales, study design, sample size and model assumptions still determine the final analysis. You can copy the reporting note as a draft, but review it before using it in a report.

Steps:

1. Click Recommendations.
2. Read Consolidated Decisions and note the workflow to which each decision refers.
3. Expand only the Recommended Statistical Analysis Path cards relevant to your question.
4. Read the Plain-language Reporting Note as a starting point for your own checked account.

Remember: The app supports initial data screening and initial normality assessment. It does not replace a full research analysis, subject-matter judgement, or independent statistical validation where required.

11. Try optional imputation carefully (05:22)

Imputation is optional. Open Imputation, choose a method, and click Run imputation. Here we demonstrate Mean only; it is not a recommendation for every study. Read the Imputation Report, then use Export imputed CSV to save the separate derived dataset. Use imputed data for screening switches the working dataset to that copy. Restore original data switches back. For important or substantial missingness, seek a justified missing-data strategy rather than treating simple replacement as a universal solution.

Steps:

1. Click Imputation, or find Optional Imputation in the left panel.
2. Choose a method appropriate for the purpose; Mean is used only as a demonstration here.
3. Click Run imputation and inspect the Imputation Report.
4. Use Export imputed CSV to save a separate derived dataset.
5. Use imputed data for screening applies the derived copy to screening. Restore original data returns to the original imported dataset.

Remember: Optional Imputation in the left panel also has icon-labelled Method and hot-deck Seed settings. Creating an imputed dataset and using it for screening are separate actions. These are single-imputation methods, not multiple imputation or full-information model estimation; document their limitations.

12. Export and document your decisions (05:59)

Before leaving, save the outputs you need. Use Export screened CSV on the Univar., Multivar., or Combined Outliers page that matches your chosen workflow. Export cases CSV saves the case-review record; Export matrix CSV saves correlations; and Export normality CSV saves the normality outputs. Keep the original CSV, your screened or imputed copies, and a note of the selected variables, cut-offs and exclusions. Refreshing or closing the app clears its working data, so download your files before you finish.

Steps:

1. Choose the Univar. Outliers, Multivar. Outliers or Combined Outliers workflow you intend to report.
2. Click Export screened CSV on that page to save its screened dataset.
3. Save Export cases CSV and Export matrix CSV when relevant; these are different outputs.
4. On Normality, use Export normality CSV. On Imputation, use Export imputed CSV for the derived data.
5. Record your variables, rules, missing-value handling, exclusions and working-data basis alongside the files.

Remember: Do not overwrite the original CSV. Closing or refreshing clears the app's working dataset, but it does not erase files you have downloaded. The optional AI assistant is not required for this workflow; do not put confidential information into its questions.