Basics of Statistics and Forecasting Webinar

SAKSHAM: IL6 Webinar for Logistics

• This webinar focused on forecasting and sampling techniques.

Basics of Statistics

• Types of Data:
  • Numerical (discrete and continuous)
  • Categorical (nominal and ordinal)
• Descriptive Statistics:
  • Mean, median, mode
  • Variance, standard deviation
  • Dependent and independent variables
  • Correlation

Forecasting Needs

• Crucial decisions requiring forecasts:
  • Market selection
  • Product planning
  • Hiring needs
  • Purchases
  • Production volumes

Forecasting Characteristics

• Forecasts are often imperfect.
• Aggregate data tends to be more accurate than individual item data.
• Forecasts tend to be more accurate for shorter time periods.

Forecasting Steps

• Define what needs forecasting.
  • Detail level, analysis units, time horizon
• Identify and assess available data.
• Select and evaluate a forecasting model.
  • Cost, ease of use, and accuracy
• Generate the forecast.
• Track forecast accuracy over time.

Forecasting Models

• Qualitative:
  • Subjective forecasts, generated by the forecaster
• Quantitative (Statistical):
  • Mathematical models generate the forecasts

Qualitative Forecasting Models

• Executive Opinion:
  • Managers meet and collaborate to produce a forecast.
  • Strengths: good for strategic or new product forecasting
  • Weaknesses: single person's opinion can dominate
• Market Research:
  • Surveys and interviews to understand customer preferences.
  • Strengths: good for understanding customer preferences
  • Weaknesses: can be difficult to create a good questionnaire
• Delphi Method:
  • Develop a consensus among a group of experts.
  • Strengths: great for long term product forecasting, technological changes, and advances.
  • Weaknesses: time consuming to develop

Statistical Forecasting Models

• Time Series Models:
  • Assumes future patterns follow past patterns and uses past trends.
  • Includes: single variable models, machine learning models
• Causal Models:
  • Explores cause-and-effect relationships.
  • Uses indicators to predict the future, such as independent/feature variables.
  • Examples include price predictions and appliance sales.

Time Series Data Composition

• Data is a combination of historic patterns and random variation.
• Historic pattern may include:
  • Level (long-term average)
  • Trend (long-term movement)
  • Seasonality (regular repeating patterns)
  • Cycle (repeating fluctuations)

Time Series Patterns

• Level (Horizontal): Data is relatively constant around a mean.
• Trend: Data is progressively increasing or decreasing.
• Seasonality: Repeating pattern within a defined time period (e.g., quarters).
• Cycle: Fluctuation over longer periods.

Level Forecasting Methods

• Naïve
• Simple Mean
• Moving Average
• Weighted Moving Average
• Exponential Smoothing

Sampling Techniques

• Population: A collection for which an inference is made.
• Sampling Units: Nonoverlapping selections of the population that covers the entire population.
• Probability Sampling:
  • Each member has a known chance of being selected.
  • Includes random, systematic, and stratified sampling
• Non-Probability Sampling:
  • Members selected non-randomly.
  • Includes convenience, judgment, quota, and snowball sampling

Sampling for populations

• Study Population: Set of people from the population to be studied.
  • Defined characteristics (e.g., age, location).
• Target Population: The larger group an inference or conclusion will be drawn from.

Simple Random Sampling

• Each member has an equal chance of being selected from the population.

Systematic Random Sampling

• Every Nth member from a list is selected.

Stratified Random Sampling

• Population divided into strata (subgroups).
• Random samples from each stratum.

Cluster Sampling

• Population divided into clusters.
• Random selection of clusters.
• All members of selected clusters are examined.

Multistage Sampling

• Multiple stages of random sampling
• Useful when a complete list is unavailable
• Costs are potentially lower, or sampling is more efficient.

Multiphase Sampling

• Collecting information from the whole sample in stages (different information per stage).

Convenience Sampling

• Using readily available participants.
• Non-probability method.

Judgment or Purposive Sampling

• Subjectively chosen respondents.
• Non-probability method.

Quota Sampling

• Segmenting population into groups representing proportions in the population.
• Convenience sampling used to select the required numbers in each stratum.
• Non-probability method.

Snowball Sampling

• Finding participants in a potentially hard-to-reach population using referrals.
• Non-probability method.

Forecast Accuracy

• Error: Difference between the actual value and the forecast value
• MAD (Mean Absolute Deviation): Average absolute forecasting errors.
• MSE (Mean Square Error): Measures the average squared difference between forecasted and observed values.
• Tracking Signal: Tracks data over time to identify forecasting issues