Data types & formats
Furthermore we assume that each column holds the responses to a specific question. The header should be first row of your file, as we do not support multi-row headers at the moment. See example table below.
| Question 1 | Question 2 | Question 3 |
|---|---|---|
| Answer of participant 1 to Q1 | Answer of participant 1 to Q2 | Answer of participant 1 to Q3 |
| Answer of participant 2 to Q1 | Answer of participant 2 to Q2 | Answer of participant 2 to Q3 |
| Answer of participant 3 to Q1 | Answer of participant 3 to Q2 | Answer of participant 3 to Q3 |
Currently you can upload your data in one of the following formats:
- Excel files (.xls, .xlsx)
- Tab delimited or comma-separated .csv file
Below you will find a list of variable types we automatically detect, parse and visualize for you.
- Categorical:Visualized as pie charts. Note, if a numeric column has less than 5 unique and consecutive values (e.g. 1, 2, 3, 4, 5), we will turn these automatically into categorical/score type and visualize it as a pie chart.
- Multiple choice:You can use any of the following separators to convey to CrowdPrisma that your collected values are from a multiple-choice field: ";", "|", "&", "\t", ",". For example, "Marketing & Business Development" will be mapped to "Marketing" and "Business Development" and visualized as two separate slices of the resulting pie-chart.
- Date:Detected automatically, given you conform to the 10 most used date formats of the web (chances are you will, so relax). We also automatically detect the range of your dates and visualize them accordingly in a histogram like line plot.
- Numeric:Visualized as a histogram like line plot.
- Text:Detected automatically and analyzed by the CrowdPrisma TextEngine to find topics and assign each response to them.
No you can't. However, if these are repeated surveys taken at different time points, you might consider stacking them on top of each other and adding a simple indicator column to denote the order of them. This indicator will be visualized as a pie-chart and will allow you to quickly switch back and forth between the different datasets.
Prisma Dashboard
You can filter every chart in the Prisma Dashboard.
- Pie charts: Click on any slice (or several) to define a subset of your data. Importantly, you can do this across as many questions as you like. You can either reset the filters by clicking the slices again, or by clicking on the refresh icon above the chart. There is also a global refresh button at the very top of the dashboard, which will reset all previously set filters. Alternatively, if you really want to start over you can simply refresh the page too.
- Line charts: Click and drag your pointer to define a selection range on any line chart (numeric or date). Then, you can adjust your selection on either side using the handles, or by dragging the entire selection left or right. As you do this, all other charts will update too automatically to reflect your chosen subset. Resetting these selections work exactly as with pie charts: local refresh buttons, global refresh button, page refresh all work.
- GroupFinder: Click and drag your pointer to define a selection area. Then, you can either move this selection around by dragging it, or change it in any direction by dragging either side of it. Cancelling a selection works by either clicking next to your chosen selection, using the global reset button at the top or by refreshing the page.
- Topics chart: Similarly to pie charts you can select any number of topics to filter your dataset. Importantly however, the TextEngine part of the dashboard will always show a single text column from your dataset. Therefore, you need to make sure that you selected the one you want to analyze at the top. Once you have done that, the topic pie chart works like any other one. Well, almost. Since a single response can be (and often is) mapped to multiple topics, this pie chart visualizes a one-to-many relationship, like the multiple-choice pie-charts.
- Sentiment chart: Click on any of the sentiments (Negative, Neutral, Positive) to define a selection. Note, that unlike topics, sentiment is a one-to-one mapping, ie. each response is mapped to a single sentiment. This is why the topic-sentiment relationship is never entirely clear-cut, only indicative. Resetting this selection work exactly as with pie charts: local refresh buttons, global refresh button, page refresh all work.
- Text Search: Type in any word into the search bar above the responses to define a new text-search filter. You can use multiple words if you separate them by "|". Don't forget the "Or | And" button on the right of the search bar. If it is switched to "And" then all your search terms separated by "|" will need to be matched in any given response for it to be surfaced. You can reset the text-search filter by simply deleting the search query in the input box.
The TextEngine of CrowdPrisma uses state of the art Large Language Models (LLM) to extract coherent themes and topics from each
of your dataset's text columns. A theme is a collection of topics that belong together. This theme & topic hierarchy speeds up the interpretation of
text responses dramatically and allows you to reason about them on two separate levels, matching the depth of your analysis.
Once the themes and topics are extracted, the TextEngine accurately assigns sentences (or response fragments) to theme - topic pairs while ensuring
the LLM is not hallucinating .
To achieve this, it will read all your responses multiple times, carefully assigning each response individually to one (or several) of the themes and topics.
Furthermore, it will also assess the general sentiment of the whole response and classify it as either positive, neutral or negative.
The result of the topic modelling is a pie chart that represents a one to many mapping of themes and topics to response. This is because each response can be mapped to multiple theme - topic pairs. To
make the job of the TextEngine even harder, these assignments can overlap too. For example: The sentence "Customers keep complaining about the bad design
of our website.", can (and should) be simultaneously mapped to both "Customer dissatisfaction" and "UX improvement" topics, and these topics naturally belong to different themes. Note that the extracted
supporting quotes for these assignments will be the whole sentence in both cases, resulting in an overlapping assignment - a unique feature of the CrowdPrisma TextEngine.
Note, since on the theme & topic level there is a one to many relationship, but sentiments are defined per response (one to one relationship), the
correspondence between sentiment and themes - topic pairs is never exact.
The GroupFinder is designed to give you an overview of your data in a single scatter chart. Each dot represent one of your respondents and
their distance from one another represents how similar they are. If two dots/respondents are close to each other, it means they replied to your
questions very similarly. Amazingly, this similarity is captured across all question types: text, categorical, multiple-choice, numeric and date.
Furthermore, you can color this plot by any of your other charts above. This makes it easy to spot patterns in your data. For example, if coloring by your first
question splits the chart into three sections, it means that this question is a key driver in separating your users / respondents into different groups.
The Autopilot is a unique feature of CrowdPrisma that allows you to immediately discover distinct groups of your respondents.
It works by automatically segmenting scatter plot of GroupFinder and looking for clusters of respondents who think and answer alike
across all your questions, be it multiple-choice or text.
You can color the GroupFinder plot with Autopilot's group colors by selecting it at the "Color by" drop down (it is at the very bottom).
Alternatively, you can click on any or a combination of the Autopilot groups to see how all the other charts (including the TextEngine)
updates.
This way it becomes easy to see what's driving the different subgroups and along which questions they differ. This becomes especially powerful of
when you start to look at the topic chart(s) to see what certain groups talk about mostly.
When you define two groups we run a number of checks before comparing them.
- We ensure both groups have at least 30 samples in them.
- We exclude any chart from the comparison that was used to define your groups.
- We only compare themes and topics if they are from the same open ended question, as comparing across different questions wouldn't make sense.
Then we perform Person's Chi-square test on each chart's data by comparing the distribution of levels (in case of a categorical pie chart) between Group A and Group B.
If the resulting p-value is lower than 0.01 we include the chart in the comparison results.
Finally, for each chart (variable) that passed the statistical test, we calculate and return the differences between the two groups.
Go to your Analyses home page and click the button under any of your dashboards to get a unique URL to your dashboard.
Conveniently, we will put this URL on on your clipboard so all you need to do is to paste it (Ctrl/Cmd + V ) into your email, tweet or presentation.
Beware, anyone in the world will be able to access your dashboard with this link. Share it with only those you intend to share your data / project with.
Note, if you generated the URL from within a dashboard (by clicking on the share icon in the top right corner), the whole state of the dashboard will be capture too. This means,
all your applied filters (except text search filters) will be loaded when the URL is opened by a recipient.
Accounts & quotas
Each month a set amount of words will be added to your quota. See the pricing page for exact numbers. For convenience
we will always display your remaining quota also in A4 pages too under your profile.
When you start a new analysis with text columns, we will calculate the number of words in your data that we will have to read and analyze for you.
If you decide to proceed with the analysis, this word count will be subtracted from your quota once the analysis is complete and ready for you to check out.
If you run out of your monthly quota, you can always top up at any time.
Don't worry, if you have unused quota left at the end of a month! We'll roll it forward for you so next month you will have that much more to work with.
Short answer is: as many as you like if you are on a paid tier (Base or Pro), if you don't use free-tex questions.
If you have open ended questions in your surveys, then as many as your quota allows. If you ever run out of your quota between your monthly billing cycles
you can always top up.
You can use almost all of our features with the free tier and run up to five analyses!
We will also allow you to upload and analyze 20 A4 pages of text within those analyses so you get a feel of the TextEngine before you commit to signing up to our paid tiers.
Unfortunately we cannot store your dashboards any longer if you downgrade to Free tier as this would be uneconomical for us.
Therefore, the dashboards which you created on a paid tier will be given a 60 days grace period and then deleted. However if you choose to reactivate your subscription within that period
we will not archive them and you will have them back in your "Analyses" tab.
Security & compliance
Access is to your data is fully restricted to you and your team. We do not access your data for any reason, unless you instruct us to do so for support purposes.
While our servers are currently located in the EU to comply with GDPR, we have customers from all over the world, therefore we will most likely increase our server locations in the future to better serve our customers.
We do use third party AI models to analyze your text responses, however we self-host these in the EU to comply with GDPR.
Please read carefully our Terms & Conditions.
Although we could end up storing copyrighted or personally identifiable data, the ultimate responsibility lies with you, the owner of that data.
This becomes especially salient if you choose to share your dashboard with a shareable link. Anyone in the world with said link can then access your dashboard and
therefore the data it was produced from.
If anyone files a claim against to us about one of your dashboards' content for either exposing PII or copyright infringement, we will investigate it and let you know of our
findings so you can respond. As set out in our Terms and Conditions, we reserve the right to remove your dashboard from our system in cases where we deem the claimant is right.
All your survey data (and derivative analysis files) are stored with the highest grade encryption in our secure cloud environment. This means, unless you share a dashboard, no one else
can access it without your account credentials.
No we are not. We are running pre-trained models on your data so we can analyze the text responses, but we are not using your data (text or other) to train any further models.