Data Exploration — descriptive analytics on demand.
The platform’s “show me what’s in this variable” tool. Pick one or more columns and get back the appropriate descriptive output for each — means and histograms for continuous variables, counts and frequency tables for categorical ones. The right summary is chosen automatically.
The business question it answers
“Before we model anything, what does the data actually look like?” — and just as often, “Quickly profile this variable for me.”
Three common ways to use it:
- Sanity check at kickoff. “What’s the response rate by region? How many people answered the brand-awareness question? Is age skewed young?” Catches data-quality issues before any modelling effort is wasted on them.
- Filter narration. “Of the under-35 respondents, what does the income distribution look like?” The exploration runs against your active chat filter, so you can see exactly who you’re talking about before drawing conclusions.
- Stakeholder asks. “The CMO wants the top 3 reasons people gave for not subscribing — pull the frequency table.” The chart and table land in chat in seconds.
How the methodology works
Data Exploration doesn’t apply a one-size-fits-all summary. It detects each column’s measurement level — using SPSS variable_measure metadata embedded in the source file when available, or falling back to pandas dtype inference — and selects the right summary accordingly.
- Continuous / scale variables. Count, mean, standard deviation, the five-number summary (min, 25th, median, 75th, max) and a histogram with KDE overlay.
- Categorical variables (nominal / ordinal). Count, unique-count, top 10 value frequencies, mode and a bar chart of category frequencies.
- Filter-aware. When you apply filters in the chat, they are applied to the dataset before the summary is computed — statistics and charts reflect only the filtered population.
This matters because computing a mean on a coded categorical variable (e.g. 1 = Male, 2 = Female) produces a meaningless number. The type-aware approach avoids this class of error automatically.
What it does not do
- No modelling.Purely descriptive. It doesn’t fit regressions, run significance tests, or compute correlations.
- No subgroup fan-out. To compare groups, ask twice with different filters, or use a modelling tool (Driver Analysis, Segmentation) when the question is comparative.
- No minimum sample size. Because no model is being fit, even small slices are reported. The n-count is always shown so you can judge reliability.
- My team currently does this in Excel and it takes hours.
- We want a tool people can self-serve from, not a report we get emailed.
- Researchers want to slice the data themselves before commissioning the heavy analysis.
Comes with the platform.
Data Exploration is built into every Crowdmines project. Run a pricing study, a driver analysis or a segmentation, and you can profile any variable in the same chat — no separate setup, no extra licence.
From a question to a chart in six steps.
Here's what you experience in the chat, from start to finish.
How Crowdmines compares to Excel pivot tables, SPSS Frequencies and survey dashboards.
Pulling frequency tables, histograms and summary statistics is the most common research task and the one most often done with the oldest tools — Excel pivot tables, SPSS Frequencies, R’s summary(), or the built-in dashboards in survey platforms. It’s not glamorous, but it’s where every project starts and where a surprising amount of analyst time gets burned.
| Capability | Traditional (Excel / SPSS) | Survey-platform dashboards | Crowdmines |
|---|---|---|---|
| Setup effort | Open the file, navigate menus or write syntax, pick variables one at a time | Log into the survey platform, navigate to the reporting tab | Type “show me the distribution of X” in the chat |
| Variable-type awareness | Analyst must know not to compute a mean on a categorical variable | Platforms usually get this right for their own data | Automatic — reads SPSS metadata or infers from dtype |
| Filtering | Manually filter in the data view or add syntax | Point-and-click filter builder | Natural language — “only women under 35” |
| Multiple variables at once | Run the command/menu per variable, combine outputs manually | Some platforms support multi-variable views | “Profile Q5, Q7, and Q12” — all returned together |
| Transition to modelling | Close this tool, open a different one, re-import data | Limited to what the platform offers | Same chat — “now run a driver analysis on NPS” |
| Analyst dependency | Requires someone who knows the tool (SPSS syntax, Excel pivots) | Researcher can self-serve but only within the platform's UI | Anyone who can describe the question |
| Speed | Minutes per variable (menu navigation, waiting for SPSS to compute) | Seconds within the platform | Seconds, for any number of variables |
The workhorse every project starts with — and keeps returning to.
Frequencies, histograms, summary stats. Filter as conversation. The right summary for the right variable, every time — no SPSS syntax, no Excel pivots, no analyst ticket queue.