Your First Data Analysis Project: One Question, a Checkable Answer

For your first data analysis project, choose one question that a small table can answer, keep the original data, check the rows you will count, and write a short report with the result and its limits. Finish when someone else can follow your calculation. A collection of charts without an answer leaves the project unfinished.

The example here asks which of two fictional community workshops recorded more attendees. You can complete it in a spreadsheet with sorting, filtering, and simple counts. It includes a duplicate booking and an unknown attendance value, so you have to decide what your numbers mean before presenting them. This is a descriptive practice exercise, not evidence of job readiness or a lesson in statistical inference.

Kognivu can turn the learning goal and your available time into daily quests. Keep the spreadsheet, cleaning decisions, and calculations in your own files. Plan a first data analysis project in Kognivu if you need help scheduling the work, then check the suggested tasks against the actual question and unfinished outputs.

Write the question before finding a bigger dataset

Start with a question that names the measure and comparison:

In this practice table, how many bookings have recorded attendance for the drawing workshop and the pottery workshop, and what remains unknown?

Each booking is for one place. The table covers only these two sessions. You are counting recorded attendance among bookings, not unique people across events, walk-in visitors, satisfaction, or all workshops the organiser has ever run.

That scope matters. A booking table cannot tell you whether everyone enjoyed the class. It also cannot establish which workshop will be more popular next month. Write those exclusions next to the question before they turn into extra tabs.

Use a small public teaching dataset when you move beyond this example, and read its documentation and reuse terms. Data Carpentry's spreadsheet lesson provides deliberately messy teaching data and exercises. It gives you a real place to practise data organisation without needing a private company export.

If you are following an assignment, its requirements take priority. For a personal practice project, postpone a dashboard, multiple joined files, predictive models, and a portfolio website until the basic answer is correct.

Keep the raw table and define what one row means

Copy this fictional table into a sheet called raw. The quote marks around one category make its trailing space visible here; the stored value inside those marks is drawing .

booking_id	workshop	attended
B01	drawing	yes
B02	drawing	no
B03	"drawing "	yes
B04	pottery	yes
B05	pottery	no
B06	pottery	unknown
B04	pottery	yes

Make a separate working copy called clean. Keep raw unchanged so you can explain the changes later. For this invented exercise, the data contract says one row represents one booking and booking_id should be unique. It also says unknown means attendance was not recorded; it does not mean the person stayed away.

For a real dataset, look for those definitions in its documentation or ask the data owner. Repeated names or identical amounts do not, by themselves, prove that rows are duplicates. Two legitimate bookings can look similar.

Data Carpentry's formatting guidance recommends a consistent table structure with a header, one observation per row, and one variable per column. Keep totals and explanatory notes outside the rows you will analyse. Put the question, field meanings, and missing-value convention in a separate notes sheet or readme.

Clean the values that affect this answer

Check the booking IDs and the distinct values in each column. In a spreadsheet, sorting or filtering lets you inspect this seven-row example without a complicated workflow. Select the whole table when sorting so the ID, workshop, and attendance stay together.

Record these decisions:

Issue	Decision for this exercise	Effect
B04 appears twice with identical values	Keep one copy, because the supplied contract identifies B04 as one booking	Seven raw rows become six bookings
drawing has a trailing space	Standardise it to drawing	B03 belongs to the intended drawing category
B06 has unknown attendance	Preserve the unknown value and count it separately	It remains a booking but cannot be counted as yes or no

For conflicting duplicate IDs, stop and investigate rather than keeping whichever row happens to come first. If B04 said yes in one row and no in another, this exercise would need a resolution rule or a clear unresolved case before you could give the same answer.

Missing values need their own meaning. Data Carpentry's common-mistakes lesson distinguishes missing observations from recorded zeros and recommends a consistent missing-value convention. Here, replacing unknown with no would invent an absence that the table does not establish.

Your change log can be plain text:

Input: 7 rows, including two identical copies of booking B04.
Removed: 1 repeated B04 row under the supplied unique-booking rule.
Standardised: B03 workshop value from "drawing " to "drawing".
Preserved: B06 attendance remains unknown.
Output: 6 bookings; attendance recorded for 5, unknown for 1.

Calculate the result, then check the totals

Create a summary from clean. Count the yes, no, and unknown values for each workshop. A pivot table or filtered counts can do this; choose the method you can explain and verify. Exclude the header from the data rows and check that your selected range contains all six cleaned bookings.

The expected summary is:

Workshop	Bookings	Recorded yes	Recorded no	Unknown
drawing	3	2	1	0
pottery	3	1	1	1
Total	6	3	2	1

Check the arithmetic before formatting the report: 3 + 3 = 6 bookings, and 3 yes + 2 no + 1 unknown = 6. For each workshop, the attendance categories must sum to its booking count.

Now distinguish two possible denominators. Drawing has two recorded attendees among three bookings, or 2/3. Pottery has one recorded attendee among three bookings, or 1/3, with one booking's attendance unknown. Those are recorded-yes shares of all bookings. They are not complete attendance rates when a value is missing.

If you restrict the calculation to bookings with known attendance, the figures become 2/3 for drawing and 1/2 for pottery. The pottery denominator changed from three to two. Label that choice explicitly; do not silently remove the unknown row and present the new percentage as though nothing changed.

The direct answer is small but defensible: drawing has two recorded attendees and pottery has one. Pottery also has one unresolved attendance value. If that value is later confirmed as yes, both workshops would have two attendees. This table does not support a confident claim that drawing is generally more popular.

Make one report someone can check

Use the summary table as the main display. A chart is optional for two categories. If you add one, show recorded yes, recorded no, and unknown separately; do not turn the unknown booking into a confirmed no-show through colour or a label.

Write the report around the question rather than around the software features you used:

Question: Which workshop has more recorded attendees in this table?
Result: Drawing has 2 recorded attendees; pottery has 1.
Coverage: 6 unique bookings for two fictional sessions; 5 attendance values known.
Preparation: Removed one contract-confirmed duplicate and trimmed one category value.
Limit: Pottery has 1 unknown attendance value; confirming it as yes would make the counts equal.
Next action: Resolve that attendance value before using the result to choose a workshop.

Keep the raw input, cleaned table, calculation or pivot setup, change log, and report together. If you used formulas, keep them visible in the working file. Save a version you can reopen. Manual spreadsheet edits need documentation; a polished sheet alone does not make the process reproducible.

Try a useful check in a separate copy: change B06 to yes. The pottery summary should become two yes, one no, and zero unknown; total bookings should remain six. Restore the original working copy afterward. This checks that your summary responds to a changed value without losing a booking.

Ask a reader to trace one number back to its rows and explain the unknown value. If they cannot, repair the notes or display before adding another chart.

Give the remaining work a finish condition

If you have six half-hour blocks available, use them as a starting estimate: one for the question and field definitions, two for cleaning and its log, one for the summary checks, one for the report, and one to reopen the files and get feedback. That totals three planned hours. Repeat or extend a block when its output is not ready; it is a time budget, not a guarantee of completion.

After each block, leave the next concrete task. “Check why the workshop totals sum to seven” is useful. “Learn more Excel” does not identify the problem you need to solve.

If you lose two blocks, keep the cleaned data, checks, and report. Drop the optional chart or move the deadline before cutting the validation. For a larger learning programme, the data analyst roadmap covers a broader sequence; this project supplies one small piece of practice. If you want to turn a specification into code instead, the first Python project guide addresses that different task.

Kognivu can help you review progress, edit your goal or schedule, and ask Coach for planning help. It does not perform this spreadsheet analysis or automatically reschedule missed work. Build my first analysis goal in Kognivu with the question, current files, finish condition, and time you can protect. Review the generated path and first quest against the outputs you still need.