FISHER function in Excel with examples of its work

FISHER function performs the Fisher transformation for the return of the arguments X. This transformation builds a function that has a normal, not asymmetric distribution. The FISHER function is used to test the hypothesis using the correlation coefficient.



Description of the FISHER function in Excel

When working with this function, it is necessary to set the value of the variable. Immediately it should be noted that there are some situations in which this function will not produce results. This is possible if the variable is:

  • is not a number. In such a situation, the FISHER function will return the error value #VALUE!;
  • is either less than -1 or greater than 1. In this case, the FISHER function will return the #NUM! error value.

The equation that is used for the mathematical description of the FISHER function has the form:

Z '= 1/2 * ln (1 + x) / (1-x)

Consider the use of this function in 3-x specific examples.



Evaluation of the relationship of profit and cost function FISHER

Example 1. Using data on the activity of commercial organizations, it is required to make an estimate of the relationship between the profit Y (million) and the cost X (million) used for product development (given in Table 1).

Table 1 - Baseline:

NoXY
1210 000 00095 000 000
21 068 000 00076 000 000
31 005 000 00078 000 000
4610 000 00089 000 000
5768 000 00077 000 000
6799 000 00085 000 000

The scheme for solving such problems is as follows:

  1. Calculate the linear correlation coefficient rxy.
  2. The significance of the linear correlation coefficient is verified on the basis of Student's t-criterion. In this case, the hypothesis about the equality of the correlation coefficient to zero is put forward and verified. When testing this hypothesis, t-statistics is used. If the hypothesis is confirmed, the t-statistic has a Student's distribution. If the calculated value of tp > tcr , then the hypothesis is rejected, which indicates the significance of the linear correlation coefficient, and hence the statistical significance of the relationship between X and Y.
  3. The interval estimate for a statistically significant linear correlation coefficient is determined.
  4. The interval estimate for the linear correlation coefficient is determined based on the inverse Fisher z-transform.
  5. The standard error of the linear correlation coefficient is calculated.

The results of solving this problem with the functions used in the Excel package are shown in Figure 1.

results functions used.

Figure 1 - Example of calculations.

Item numberName of the indicatorCalculation formula
1Correlation coefficient=CORREL(B2:B7,C2:C7)
2Estimated t-criterion t=ABS(C8)/SQRT(1-POWER(C8,2))*SQRT(6-2)
3The table value of the t-criterion trh=TINV(0.05,4)
4Tabular value standard. normal Distr. zy=NORMSINV((0.95+1)/2)
5Fisher transform value z ’=FISHER(C8)
6Left interval estimate for z=C12-C11*SQRT(1/(6-3))
7Right interval estimate for z=C12+C11*SQRT(1/(6-3))
8Left interval estimate for rxy=FISHERINV(C13)
9Right interval estimate for rxy=FISHERINV(C14)
10Standard deviation for rxy=SQRT((1-C8^2)/4)

Thus, with a probability of 0.95, the linear correlation coefficient lies in the interval from (–0.386) to (–0.990) with a standard error of 0.205.

Checking the statistical significance of the regression by the FINV function

Example 2. To verify the statistical significance of the multiple regression equation using Fisher's F-test, to draw conclusions.

To test the significance of the equation as a whole, we hypothesize H0 about the statistical insignificance of the coefficient of determination and its opposite hypothesis H1 about the statistical significance of the coefficient of determination:

H0 : R2 = 0;

H1 : R2 ≠ 0.

We test the hypotheses using the Fisher F-criterion. The indicators are shown in table 2.

Table 2 - Baseline

IndicatorSSMSFcalc
Regression454,814227,4077,075
Remainder1607,01432,14
Total2061,828-

For this we use the function in Excel:

=FINV(α,p,np-1)

Where:

  • α is the probability associated with a given distribution;
  • p and n are the numerator and denominator of the degrees of freedom, respectively.

Knowing that α = 0.05, p = 2, and n = 53, we obtain the following value for Fcrit (see Figure 2).

FINV used.

Figure 2 - Example of calculations.

Thus, we can say that Fcalc > Fcrit. As a result, the hypothesis H1 about the statistical significance of the coefficient of determination is accepted .

Calculation of the correlation index in Excel

Example 3. Using data from 23 companies on: X - the price of product A, thousand; Y - profit of commercial enterprise, study of their dependence is made. Evaluation of the regression model gave the following: ∑(yi-yx)2 = 50,000; ∑(yi-yср)2 = 130000. What correlation indicator can be determined from this data? Calculate the value of the correlation indicator and, using the Fisher criterion, draw a conclusion about the quality of the regression model.

Define Fcrit from the expression:

Fcalc=R2/23*(1-R2)

where R is the coefficient of determination equal to 0.67.

Thus, the calculated value of Fcalc = 46.

To determine Fcrit, use the Fisher distribution (see Figure 3).

Fisher distribution.

Figure 3 - Example of calculations.

Download examples functions FISHER in Excel

Thus, the resulting estimate of the regression equation is reliable.


en ru