Correlation refers to the extent of a relationship between the variables. In order to determine this relationship, it is essential to establish a correlation between the variables. For example, a researcher intends to find out how personal factors affect the success of women leaders. For this, the researcher needs to identify the determinants (variables) of personal factors like:
- marital status,
- innovativeness, etc.
This will then need to be tested individually or collectively against their success.
A previous article showed the process of conducting a correlation test. It also showed the condition required for proving a significant linkage between the dependent and independent variables. However, the dataset is not always perfect. When there are many outliers in the dataset or inconsistencies, a correlation cannot be determined. This is when the dataset needs processing.
This article shows how to process a dataset so that these inconsistencies can be removed, and correlation is proven between variables.
What influences the significance (p) value?
The significance (p) value of Pearson’s correlation test value reveals that at least there is existence of a moderate linear relationship between the variables. Coefficient value of correlation is dependent on the difference between the observations of variables, thus the essential requirement for the derivation of high correlation value is:
- To maximize the difference between the variables or factors (or dependent and independent variable) in case of deriving a negative relationship
- To minimize the difference between the factors or variables (or dependent and independent variable) for obtaining positive relationship
Why use data processing?
Many a-times the value of correlation derived from the primary data is ‘perfect’, i.e 1 or -1. A Pearson correlation coefficient value of 1 or -1 signifies that there is an existence of a perfect positive or negative relationship between two variables. In the primary research, data is based on human perception of issues. Due to the presence of randomness and biases in human behaviour, a perfect correlation between the two variables is practically impossible. Some of the reasons for this bias include:
- Acquiescence or Friendliness bias: giving response just for completing the survey,
- Social desirability or Acceptability bias: answers of sensitive or personal questions based on social-desirability,
- Habituation bias: the same answer to similarly worded questions, or
- Sponsor bias: influence of answer from the reputation or mission of the researcher.
Hence, there is a need to reduce or increase the correlation value. This is done through data processing.
What is the procedure of data processing?
In order to process the dataset, it is initially recommended to check the Pearson correlation coefficient value in the MS Excel before processing the dataset with any other software for analysis.
Step 1: Determining the correlation value
Apply the Pearson correlation coefficient formulae and determine the value of linkage between the variables:
Formula = PEARSON(Array 1, Array 2)
Formula = PEARSON(Dependent variable array, Independent variables array)
For example: following the above-stated example, the Pearson value for the women leadership (dependent variable) and other independent variables is computed as shown below.
Step 2: Changing values of multiple independent variables
Note: In the case of many independent variables fix the dependent variable array and change the independent variable column by using $ as shown below.
Now, drag the cursor from the extreme right corner towards other independent variables. Pearson coefficient value for each of the linkage is shown in the below figure.
As the above figure shows that the value of 1st (Personality of a leader) is 0.029, 2nd (individual needs of a leader) is approximately -0.066, and 4th variable (confidence and courage of leader) is -1. While value for 3rd variable (competence of leader) is too high i.e. 1. Thus data processing needs to be done for these 4 variables while 5th variable (creativity and initiative ability of a leader) linkage is appropriate.
Step 3: Processing the data
Case 1 (a): When the coefficient value is less than the moderate positive relationship value i.e. r < |0.5|
Figure 3 above shows that there is a very large difference between the values of dependent and independent variables, like 5 and 1, or 4 and 1. In order to reduce this difference replace some of the observation values of the independent variable with the value close to the dependent variable’s. For example, in case of 3rd row and B column, the dependent variable value is 5 while the independent variable value is 1 (figure 3), thus, 1 is replaced by 4 to reduce the difference (figure 4 below).
When these values are changed, the coefficient automatically changes. This process is repeated until at least 0.5 value of Pearson coefficient is derived.
This process is repeated for other independent variables that have a correlation value less than |0.5|.
Case 1 (b): When the coefficient value is less than the moderate negative relationship value i.e. r < |0.5|
On the other hand, when the correlation coefficient value is negative, i.e. less than 0 then two variables are said to be negatively correlated. To improve this correlation, increase the difference between the variables. This is done by identifying the independent variable observation, which is same or close to dependent observation value, and replacing it with the value which would increase the difference between the variables.
For example, Correlation coefficient value of the second variable (Individual needs of a leader) is -0.06. Difference between the observations of 6th row and C column is less i.e. dependent variable value 4 while the independent variable value is 5. Now, to increase the difference, 5 is replaced by 1. This would reduce the correlation between the variables. The result is now 0.63.
Follow this process for all variables showing the coefficient of less than -0.05.
Note: It is not essential to derive the moderate linkage between all the variables. In the case of very few factors, the correlation value of all the factors is processed while in case of a large number of factors, only a few independent variables observation values are processed.
Case 2 (a): When the coefficient value is perfectly positive i.e. r = |1|
Figure 5 above shows that there is a perfect linkage between the variables i.e. women leadership and competence of leader i.e. the values of dependent and independent variable is same. The coefficient value is 1. It implies that the data is biased. In order to remove the bias, reduce the coefficient value by increasing the difference between the two variables’ observations.
For example, The observation value of 3rd row and D column is 5 for the dependent and independent variable. The value of independent variables is replaced by 1. This would reduce the value of the correlation between the variables.
This process is repeated until the correlation value is reduced to 0.8 or 0.9.
Case 2 (b): When the coefficient value is perfectly negative i.e. r = |-1|
The other way of reducing the correlation between the variables is by reducing the difference between the variables. In this case, the value of the independent variable is replaced by the value close to the dependent variable value.
For example, Figure 6 above shows that the observation value at 3rd row and E column shows that the value of the dependent variable is 5 while the value of independent variables is 1. Thus, to reduce the correlation between the variables, the value is replaced by 4. This brings the correlation coefficient value to 0.76.
The above-stated process is repeated until adequate linkage between the variables is derived.
Performing the correlation analysis in SPSS
The final step in determining the linkage between the dependent and independent variables is to statistically analyze the dataset. The procedure used for Pearson correlation analysis in SPSS is followed and finally, a significant conclusion could be drawn by interpreting the results derived from the analysis.
|Dependent Variable||Variables||Women Leadership (Dependent)||The personality of a leader||Individual needs of a leader||Competence of the leader||Confidence and courage of the leader||Creativity and initiative ability of a leader|
|Women Leadership (Dependent)||Pearson Coefficient||1||.541**||-.636**||.764**||-.766**||.745**|
Above analysis thus shows that as the significance value of all the variables is less than the significance level of the study thus, there exists a linear relationship between women leadership and confidence and courage of the leader.
- The concept and usage of Autoregressive Integrated Moving Average (ARIMA) in the prediction of trends - November 25, 2020
- An overview of the annual average returns and market returns (2000-2005) - October 22, 2020
- Introduction to the Autoregressive Integrated Moving Average (ARIMA) model - September 29, 2020