🧩 Data Shape for Fixed-Point & Relative-Time Visualizations

🐍 From Wide → Long in Python

When your computed dataset looks like this:

fed_event_date	n_plus_1	n_plus_2	n_plus_5
2025-10-29	0.3%	0.7%	1.9%
2025-11-13	-0.2%	0.4%	1.1%

…you can reshape it into a long format that Tableau (and any tidy-data workflow) will love.

import pandas as pd

# Example input: wide format
wide = pd.DataFrame({
    "fed_event_date": ["2025-10-29", "2025-11-13"],
    "n_plus_1": [0.3, -0.2],
    "n_plus_2": [0.7, 0.4],
    "n_plus_5": [1.9, 1.1],
})

# Melt the wide columns into long form
long = wide.melt(
    id_vars=["fed_event_date"],          # columns to keep
    var_name="horizon_label",            # name of the new "dimension" column
    value_name="return"                  # name of the measure column
)

# Optionally extract numeric horizon (e.g. 1, 2, 5)
long["rel_horizon"] = long["horizon_label"].str.extract(r"(\d+)").astype(int)

print(long)

Output:

fed_event_date	horizon_label	return	rel_horizon
2025-10-29	n_plus_1	0.3	1
2025-10-29	n_plus_2	0.7	2
2025-10-29	n_plus_5	1.9	5
2025-11-13	n_plus_1	-0.2	1
2025-11-13	n_plus_2	0.4	2
2025-11-13	n_plus_5	1.1	5

Now Tableau can use:

• rel_horizon (or horizon_label) as the X-axis
• return as the Y-axis
• fed_event_date as the Color or Detail dimension.

---

1️⃣ The Guiding Principle

Data meant to be visualized should be long, not wide. Each thing you want to put on an axis should be a dimension, and each thing you want to aggregate should be a measure.

---

2️⃣ Why Wide Data Happens

Wide tables are a computational artifact, not an analytical design. They emerge naturally from:

• Vectorized calculations (shift(-1), shift(-5), etc.)
• Spreadsheet habits (one column per metric)
• Human readability (“I want to see everything for one event in one row”)

In other words: wide is for computation or inspection — not for visualization.

---

3️⃣ Why Long Is the Correct Shape for Visualization

When you pivot from wide → long, you’re really expressing a semantic truth:

“N+1,” “N+2,” “N+5,” etc. are values of a single conceptual variable: time since event.

That variable — call it horizon or rel_time — belongs on the X-axis. Your measure (like return) belongs on the Y-axis. Tableau will then:

• Aggregate return by horizon
• Draw continuous lines
• Let you filter, color, or facet by any other dimension (e.g. fed_event_date)

---

4️⃣ In Practice

fed_event_date	horizon	return
2025-10-29	N+1	0.3%
2025-10-29	N+2	0.7%
2025-10-29	N+5	1.9%
2025-11-13	N+1	-0.2%
2025-11-13	N+2	0.4%

X-axis: horizon Y-axis: return Color: fed_event_date

Every line shares a fixed origin (N=0) — perfect for event studies, cumulative performance charts, or time-relative comparisons.

---

5️⃣ When (Rarely) Wide Makes Sense

There are only a few edge cases where you might intentionally keep wide data for visualization:

• Static metric tables / KPI dashboards
• Correlation matrices (when each column is an axis of the grid)
• Parallel coordinates plots

Everything else — especially time-based or event-based visuals — belongs in long form.

---

🧠 Rule of Thumb

If you find yourself renaming columns just to plot them, your data should probably be long instead of wide.

---

🎨 Even Grids Are Long

Even visuals that look like wide data — such as heat maps — are conceptually long underneath.

🔹 What people think a heat map is

	N+1	N+2	N+3
Event A	0.2	0.5	0.8
Event B	0.1	0.4	0.7

That looks like a 2D matrix, but Tableau still wants one row per cell, not one row per event:

event	horizon	return
A	N+1	0.2
A	N+2	0.5
A	N+3	0.8
B	N+1	0.1
B	N+2	0.4
B	N+3	0.7

• event → Rows
• horizon → Columns
• return → Color

Even though it renders as a grid, it’s still a long-form dataset.

🔹 The conceptual rule

Any time you have two things you want to cross on a grid, those two are dimensions, not measures. Measures fill the cells; dimensions define the grid.

---

🔁 Converting Back: Long → Wide

Sometimes you’ll want to pivot back into a wide form — for example, when exporting Tableau results or doing additional vectorized calculations.

# Starting from the "long" dataframe above
wide_again = long.pivot(
    index="fed_event_date",       # unique key or grouping column
    columns="horizon_label",      # dimension to expand into columns
    values="return"               # measure to fill cells
).reset_index()

# Optional: remove the pivot MultiIndex that Pandas creates
wide_again.columns.name = None

print(wide_again)

Output:

fed_event_date	n_plus_1	n_plus_2	n_plus_5
2025-10-29	0.3	0.7	1.9
2025-11-13	-0.2	0.4	1.1

This restores the original computational layout, ready for analysis or export.

---

🧭 Summary

Concept	Wide Format	Long Format
Purpose	Computation / readability	Visualization / aggregation
Structure	One row per event	One row per (event, horizon)
Tool Compatibility	Spreadsheets, NumPy, SQL	Tableau, Power BI, analytics tools
Best For	Calculations	Charts, heatmaps, interactivity

---

🪄 In Plain English

The moment you decide something belongs on an axis or label — it’s a dimension. Everything left over becomes a measure.