Statistics outcomes

Statistics outcomes

1. regression analysis (linear and not linear)

   • hypothesis testing
     • Null hypothesis
       • Alternate hypothesis
     • T-test statistical significance of a variable
       • To find out is there significant difference between coefficients in regression and zero
       • made for each of coefficient
     • Normal distribution of an error
     • Two-tailed and one-tailed test
     • F-test significance of an overall model
       • At leat one coefficient is not equal to zero
       • adjusted R^2 (coefficient of determination) shows the percent of items, described by model
         • to compare models
       • check the statistical significsance of an overall model
         • by comparing
           • F statistical value from excel
             • to
               • F critical vavue from a table
       • for all coefficients
   • Multiple variables
     • Adjusted
   • single variable
     • What is the correspondance between two variables?
       • Is there such a correspondance?
   • makes hypothesises
   • Problems
     • multicolliniarity
     • functional form misspecification
     • hestroscodosticity
       • the value of an error
     • Outliers, that are destorting the regression line
       • Understand their input and reasons for that
         • consider adding additional variables to capture their influence
           • may primarilt be qualitative in nature
     • autocorellation
       • value depends in a value from a previous period
         • first order - from previous period
           • durbin watson test for first otlrder autocorellation
         • second order - form value two periods before
       • leads to unreliable tests of significance
       • almost always present at times series data
         • if a positive (negative) residual in one time period is followed by a positive (negative) residual in a next time period
           • positive autocorellation
         • positive - negative
           • negative autocorellation
       • seasoalyty
       • Durban Watson test for auticorellation
         • DW close to zero - positive autocorellation
         • DW close to four - negative autocorellation
         • DW close to 2 - no autocorellation
   • types
     • linear
     • random walk
       • with drift
       • without drift
     • exponential / logarythmic
     • quadratic
     • dependent on previous value plus value

2. Factor anslysis - many variables, fewer factors (linear combination of variables), that influence the result more

   • Principal factor analysis
     • SPSS package

3. cluster analysis

   • how many clusters to look for
   • differnt techniques gives different results
     • One of the simpiliest - nearest neighbour clustering