Date: 5th October, 2015

Session by: Dr. Saswata Ghosh. Asst. Professor, IDSK.

o Correlation analysis can be applied to continuous variables (e.g. Weight, Height,

temperature etc.) but it does not explaining causality.

o Data set (Saswata_New): Survey on peoples of West Bengal. (Sample: Ever married

Women of age: 15-49)

o First check the frequencies to ascertain that, there is no missing data i.e.

inconsistencies or irrelevance in the data. (Note: in this case, we have performed on

the variables: Age and education in single year) For missing value: percent shall not

be matched with valid percent. Cumulative Freq. is calculated based on valid

percent. Missing values are not calculated in valid percent. After that, perform

Pearson’s correlation: Correlate -> Bivariate between the same value . It shows – ve

correlation at 1% significance level but the value is very low.

o Partial: Correlate -> Bivariate -> Partial: enter two variables into variables and one

controlling variable (which we keep constant for this analysis)

o If in the output, no particular significance level is mentioned, it indicates that, result

is valid for both the level of significance i.e. 1% and 5%.

o Causality: Regression; Correlation: Degree of association.

o Correlation analysis is valid for linear. For large data set, it is assumed that, the

distribution is following normal distribution in which more or less linearity is followed.

Otherwise we need to ascertain linearity first. (Ref: ANOVA methodology)

o (Variable – mean)/ SD = Standardized value. For having clear inference, Standardize

but in this case we lose original character of the result. Recommendation is first go

for unstandardized one for explanation, if difficulty arises go for standardization.