TOUT SUR LES STATISTIQUES DESCRIPTIVES... Simplement.

Main ideas and lessons (descriptive statistics)

• Goal of descriptive statistics: Use existing data to analyze and describe it (without probabilistic prediction).

• Contrast with inferential statistics:

  • Inferential statistics uses probabilities to predict/anticipate and make decisions.
  • Descriptive statistics summarizes what is already observed.

Types of variables discussed (3 main categories)

1. Quantitative variables (often the focus in the session)

   • Discrete quantitative: values are integers (e.g., number of children).
   • Continuous quantitative: values lie on an interval; infinitely many values between bounds (e.g., seniority in years).

2. Qualitative variables

   • Nominal: categories with no ranking (e.g., single/divorced/married/widowed).
   • Ordinal: categories with an order/scale (e.g., very satisfied → unsatisfied).
   • Binary: two modalities (e.g., male/female).
   • Qualitative variables cannot be added/subtracted, but they can be counted.

3. Temporal variables (dates/times)

   • You can subtract (time/date differences), but not reliably add/multiply.

---

Practical exercise used throughout: seniority of employees (N = 80)

• The dataset conceptually represents 80 employees grouped by seniority classes (quantitative, continuous).
• From the same grouped series, the video computes and interprets:
  • Mode (via modal class)
  • Mean (average) (two methods; one emphasized as faster)
  • Median (multiple equivalent approaches: frequencies, cumulative frequencies, and graphical)
  • Quartiles (Q1 and Q3; using interpolation)
  • Deciles (example: D9)
  • Variance and standard deviation (two methods; interpretation of dispersion)

---

Methodology / instructions (step-by-step)

1) Mode (modal value for grouped quantitative data)

A. When class width/amplitude is identical

• Step 1: Identify the modal class = class with highest frequency.
• Step 2: Use the class boundaries:
  • Let modal class be ([L, U]) (lower and upper class edges)
  • Let:
    • (f_m) = frequency of modal class
    • (f_{prev}) = frequency immediately before modal class
    • (f_{next}) = frequency immediately after modal class
    • (h = U - L) = class width (amplitude)
• Step 3 (formula idea): Apply interpolated mode using differences of adjacent frequencies, multiplied by class width.
• Interpretation: The computed mode is the most typical seniority within the grouped distribution.
• Video note: Taking the midpoint works only approximately; the video uses a more precise interpolation.

B. When class widths/amplitudes are unequal

• Step 1: Compute the class amplitudes for each class.
• Step 2: Compute a corrected frequency via density:
  • density = frequency / amplitude
• Step 3: Choose a common reference amplitude (the video suggests something simple like 10) to convert densities into comparable “adjusted frequencies.”
• Step 4: Determine the modal class using the adjusted frequencies.
• Step 5: Apply the same interpolated mode method as in the equal-width case, using:
  • adjusted modal frequency
  • adjusted neighbor frequencies
  • modal class width.

---

2) Mean (average seniority) for grouped continuous data

A. Faster method emphasized (midpoints + frequencies as percentages)

• Step 1: For each class, compute the class midpoint (C): [ C = \frac{L + U}{2} ]

• Step 2: Compute frequencies (f_i) as percentages: [ f_i = \left(\frac{n_i}{N}\right)\times 100 ]

• Step 3: Compute weighted contributions (C \times f_i).

• Step 4: Sum to get the mean: [ \bar{x} = \frac{\sum (C \cdot f_i)}{100} ] (In the video’s practice: because (f_i) are already percentages, the summation leads directly to the mean after consistent scaling.)

• Step 5 (check): Ensure the percentages sum to approximately 100% (minor rounding allowed).

B. Verification method (classic grouped mean formula)

• Step 1: Compute (\sum (n_i \cdot C_i)).

• Step 2: Divide by total (N): [ \bar{x} = \frac{\sum (n_i C_i)}{N} ]

• Interpretation: The mean represents the average seniority (the video cautions about interpreting decimal values as literal fractions of a year).

---

3) Median

Core definition

• The median corresponds to the 50% cumulative position.

A. Using cumulative increasing frequencies (percentages) + linear interpolation

• Step 1: Build cumulative increasing frequency (ascending):
  • add class frequencies until reaching 100%.
• Step 2: Find where 50% lies:
  • it falls between two cumulative points (e.g., between 48.75% and 66.25%).
• Step 3: Identify the class interval containing 50%.
• Step 4: Apply linear interpolation within that class using:
  • lower bound (X_1), upper bound (X_2)
  • cumulative frequency at (X_1): (F(X_1))
  • cumulative frequency at (X_2): (F(X_2))
  • target = 50%
• Output: median in years (then interpreted as years plus a fraction of a year).

B. Graphical method

• Step 1: Plot:
  • CF ascending (cumulative increasing frequency)
  • CF descending (cumulative decreasing frequency)
• Step 2: Plot against seniority class midpoints (x-axis).
• Step 3: The intersection gives the median position.
• Step 4: Read the median from the x-axis.

C. Alternative computational method using (N/2) with counts

• Step 1: Compute (N/2) (using counts, not percentages).
• Step 2: Locate (N/2) in the cumulative count column.
• Step 3: Interpolate within the class containing (N/2).
• Interpretation: half the employees have seniority below that value and half above it.

---

4) Quartiles (cartilaginous percentages): Q1 and Q3 (and link to median)

Definitions

• Q1 = 25%
• Q2 = 50% (the median)
• Q3 = 75%
• Q4 = 100% (mentioned as corresponding to the upper data limit)

Step-by-step for Q1 and Q3 (linear interpolation)

• Step 1: Using cumulative increasing frequency:
  • locate 25% for Q1
  • locate 75% for Q3
• Step 2: Apply linear interpolation inside the identified class:
  • use class bounds (X_1, X_2) and cumulative frequencies at those bounds.
• Step 3: Convert the result to years + fraction of a year.
• Interpretation examples:
  • Q1: 25% of employees have seniority below that value.
  • Q3: 75% have seniority below that value (so 25% are above it).

---

5) Deciles

Definition

• There are 10 deciles dividing the ordered distribution into 10% steps:
  • D1 = 10%, D2 = 20%, …, D9 = 90%, D10 = 100%
• The video demonstrates one decile: D9 (90%).

Step-by-step (example: D9)

• Step 1: Locate 90% in the cumulative increasing frequency column.
• Step 2: Determine the class interval where 90% lies.
• Step 3: Use linear interpolation inside that interval between class edges and cumulative frequencies.
• Interpretation: D9 means 90% of employees have seniority below that value.

---

6) Variance and Standard Deviation

Core formulas

• Variance (grouped data concept): [ V = \frac{\sum n_i (x_i - \bar{x})^2}{N} ]

• Standard deviation: [ \sigma = \sqrt{V} ]

A. Classic table method (center minus mean)

• Step 1: Compute mean (\bar{x}).
• Step 2: For each class midpoint (C_i), compute:
  • (d_i = C_i - \bar{x})
• Step 3: Square (d_i^2).
• Step 4: Multiply by class count (weight):
  • (n_i \cdot d_i^2)
• Step 5: Sum and divide by (N) to get variance.
• Step 6: Take the square root for standard deviation.

B. Faster method idea

• Step 1: Reuse the class midpoints and frequency/proportion data.
• Step 2: Build a weighted squared-deviation column:
  • (f_i \cdot (C_i - \bar{x})^2)
• Step 3: Sum and apply variance scaling as needed.
• Step 4: Take the square root for (\sigma).

Interpretation of standard deviation (dispersion)

• Standard deviation describes how spread out seniority values are around the mean.
• The video interprets it using the band:
  • mean − σ to mean + σ
• It connects dispersion to whether there are many short-tenured and many long-tenured employees.

---

Speakers / sources featured

• Speaker: the video instructor/lecturer (only one person; no name provided in the subtitles).