Comprehensive Study Notes on Correlation, Linear Regression, and Bivariate Data Analysis

Financial Calculations and Variable Identification

The initial calculations involve specific currency values and date-related data points. A calculation is presented as $1,13 \times 1,14$ and $7442$ resulting in $= 0.6934 / 100\%$. A specific value of $2081.54\,PLN$ is recorded alongside the date $23.04.2026$. Another sequence of values includes $1.13-1.14$, $1.12$, and $1.11 = 1,6015 - 1,300$. These appear to be related to exchange rates or financial coefficients.

In the study of statistics, variables are identified by their roles in a relationship. In this dataset, the variable $X$ represents income, and the variable $Y$ represents spendings. The relationship between these two can be visually represented on a scatter plot. Different patterns are identified: a positive correlation that is not linear, and a positive linear correlation where the data points align closely to a straight line.

Introductory Example: Bonus and Job Satisfaction

A practical example is provided to illustrate the relationship between the amount of a bonus ($X$) and job satisfaction ($Y$). The sample size is denoted as $n = 25$. Specific frequency counts are tracked across different categories. For instance, at the beginning, the values are given as $X = \text{amount of bonus}$ and $Y = \text{satisfaction}$.

Indices are used to track specific data points within a frequency table, such as $n_{i1} = 10$ and terms like $n_{13} = 1$. Marginal frequencies are essential for analyzing bivariate data. The term $n_{i.}$ refers to the marginal frequency for the $X$ variable (sums across rows), while the term $n_{.j}$ refers to the marginal frequency for the $Y$ variable (sums across columns).

Parameters of Marginal and Conditional Distributions for X

Detailed distributional parameters for the $X$ variable are calculated using specific frequency tables. The data utilizes midpoints ($x_i$) and counts ($n_{i.}$). Calculations for the mean include terms such as $x_i n_{i.}$ and $x_i^2 n_{i.}$. The specific sums recorded are $\sum n_{i.} = 25$, $\sum x_i n_{i.} = 19.250$, and $\sum x_i^2 n_{i.} = 22.563$.

The arithmetic mean of $X$ is calculated using the formula $\bar{x} = \frac{1}{n} \sum x_i n_{i.}$, which for this dataset is $\bar{x} = \frac{19.25}{25} = 0.77$.

The variance is determined by $S^2(x) = \frac{1}{n} \sum x_i^2 n_{i.} - (\bar{x})^2$. Applying the numbers:$S^2(x) = \frac{22.563}{25} - (0.77)^2 = 0.310. \n\nThe standard deviation is the square root of the variance: S(x) = \sqrt{0.310} = 0.556. \n\nAdditional complex calculations involve the sum of products \sum xy$, which is recorded as$74.75$and a value of$200.52$in expanded tables. The Pearson correlation components are introduced here, showing how individual observations of$x$($1, 2, 3, 4, 5, 6, 8, 4) map to their respective frequencies.\n\n# Parameters of Marginal and Conditional Distributions for Y\n\nSimilar to variable X$, the distribution of$Y$is analyzed. The frequencies ($n_{.j}$) for values of$y$are summed to$n = 25$. The values of$y$include$2$,$3$, and others. The sums result in$\sum y_j n_{.j} = 31.5$(later adjusted or recalculated in contexts to$23.75$or specific sums like$39$and$81.5). \n\nIn one specific iteration of the data, the mean is calculated as \bar{y} = \frac{31.5}{25} = 1.26$or, using another subset of data,$\bar{y} = \frac{81.5}{25} = 3.26. \n\nThe variance for Y$is given as$S^2(y) = \frac{1}{n} \sum y_j^2 n_{.j} - (\bar{y})^2$. For the value set where$\bar{y} = 3.26$, the calculation is proceeds as$S^2(y) = \frac{300.5}{25} - (3.26)^2 = 12.02 - 10.6276 = 1.3924. \n\nThe standard deviation for Y$is calculation as$S(y) = \sqrt{1.3924} = 1.18$. In other notes, specific values such as$S(y) = 2.73$or$1.2 are mentioned based on different data subsets.\n\n# Pearson Correlation Coefficient (r)\n\nThe Pearson Correlation Coefficient, denoted as r_{xy}$or$r_{yx}, is defined specifically for linear relationships. It measures the strength and direction of the linear association between two variables. \n\nThe formula for the correlation coefficient is r_{xy} = \frac{COV(x, y)}{S(x) S(y)}$, where$COV(x, y) is the covariance. \n\nThe value of r$ranges from$-1$to$1: \n- r = 0: No correlation.\n- r > 0: Positive correlation.\n- r < 0: Negative correlation.\n\nA specific covariance calculation is shown: COV(x, y) = \frac{1}{n} \sum x_i y_j n_{ij} - \bar{x} \bar{y}$. Based on the calculated values of$\bar{x}$,$\bar{y}$, and$\sum xy$, the coefficient in the satisfaction example was determined to be$r_{yx} = 0.76. This is interpreted as a high, positive linear correlation between the amount of bonus and job satisfaction.\n\n# Correlation Ratio and Non-Linearity\n\nWhen a relationship is not linear, the correlation ratio (denoted as \eta$or$e$) is used. It is calculated using the variance of the conditional means. The formula involving the empirical variance is$S^2(y) = S^2(y_c) + S^2_{res}(y)$, where$S^2(y_c)$is the variance explained by the regression (inter-group variance) and$S^2_{res}(y) is the residual variance (within-group variance).\n\nThe correlation ratios are given by: \ne_{yx} = \sqrt{\frac{S^2(y_c)}{S^2(y)}}\ne_{xy} = \sqrt{\frac{S^2(x_c)}{S^2(x)}}\n\nTo measure the degree of curve linearity, the measure m_{yx}$is used:$m_{yx} = e^2_{yx} - r^2_{yx}$. If$m_{yx} \le 0.3$, the correlation is considered linear. If$m_{yx} > 0.3$, the correlation is not linear. In the satisfaction example,$m_{yx} = (0.75)^2 - (0.76)^2$, which results in a value close to zero ($0.01$), confirming that the correlation is linear since it is less than$0.3.\n\n# Linear Regression Modeling\n\nTheoretical regression is represented by the equation of a straight line: \hat{y} = ax + b. \n\nThe slope a$is calculated as$a = r \times \frac{S(y)}{S(x)}. \nIn the example, a = 0.76 \times \frac{1.3}{1.035} = 0.96.\n\nThe intercept b$is calculated as$b = \bar{y} - a\bar{x}. \nExample: b = 2.5 - 0.96 \times 1.15 = 1.4. \nThis gives the theoretical regression line: \hat{y} = 0.96x + 1.4. \n\nThis model can be used for prediction. For instance, if X = 1.23$, then$\hat{y} = 0.96 \times 1.23 + 1.4 = 2.58. \n\n# Coefficient of Determination\n\nThe Coefficient of Determination (R^2$) measures the proportion of variability in the dependent variable that is explained by the regression model. It ranges from$0$to$1.\n\nThe formula is R^2 = 1 - \frac{\sum(y_i - \hat{y}i)^2}{\sum(y_i - \bar{y})^2}. \nFor linear models, R^2 = r^2. \nIn the bonus/satisfaction example, R^2 = (0.76)^2 = 0.58$. This means approximately$58\% of the variability in satisfaction is explained by the bonus amount.\n\n# Advanced Examples: Sales and Invoices\n\nA dataset regarding sales representatives (X$) and computers sold ($Y$) provides further practice. For a sample size$n = 20:\n- Sums: \sum x = 44$,$\sum y = 122$,$\sum x^2 = 108$,$\sum y^2 = 826$,$\sum xy = 310.\n- Means: \bar{x} = 2.2$,$\bar{y} = 6.1.\n- Variances: S^2(x) = 0.56$,$S^2(y) = 9.09.\n- Standard Deviations: S(x) = 0.75$,$S(y) = 3.01.\n- Covariance: COV(x, y) = \frac{155}{20} - (2.2 \times 6.1) = 7.75 - 13.42 = -5.67$(Note: alternative calculations provided in the transcript suggest$COV = 2.08$or$0.92 depending on context).\n- Regression result: \hat{y} = 3.69x - 2. \n- If 20$computers are predicted ($Y=20$), the number of reps needed is$20 = 3.69x - 2 \rightarrow x = 5.96.\n- If 5$sales reps are employed ($X=5$), predicted sales are$\hat{y} = 3.69 \times 5 - 2 = 16.45.\n- A calculated R^2 = 0.8498$indicates that$84.98\% of the variability in the number of computers is explained by the number of sales representatives.\n\nA separate case regarding Account task/invoices and consumption (C$) vs income ($Y$) follows a similar structure. A regression equation for consumption is given as$\hat{C} = 250 + 0.911Y$. Under this model, if Income ($Y$) is$0$, the consumption expenditure is$250$. This relates to the economic concept of "Marginal Propensity to Consume" ($C = C_0 + cY$). If consumption is targeted at$2500$, then$2500 = 250 + 0.911Y$, resulting in an income requirement of$Y = 1372.12. \n\n# Distribution Tables with Grouped Data\n\nFor larger datasets involving grouped data or ranges (e.g., 0-2$,$2-4$, etc.), midpoint calculations are critical. A dataset with$n = 650$shows calculations for the marginals of$A and income ranges. \n\nFor the marginal of X in this group:\n- \sum x_i n{i.} = 2686\n- \bar{x} = \frac{2686}{650} = 4.132\n- S^2(x) = \frac{16926}{650} - (4.132)^2 = 26.04 - 17.07 = 8.97 \rightarrow S(x) = 2.99\n\nFor the marginal of Y$in this group ($n = 650):\n- \sum y_j n_{.j} = 5284\n- \sum y_j^2 n_{.j} = 56672\n- \bar{y} = \frac{5284}{650} = 8.129\n- S^2(y) = \frac{56672}{650} - (8.129)^2 = 87.18 - 66.08 = 21.10 \rightarrow S(y) = 4.59\n\nConditional distributions are also analyzed, such as the conditional distribution of X$for specific ranges of$Y$(e.g., for individuals having between$4$-$8$months). The mean of$X$for that specific group is calculated as$3.31$and the standard deviation as$1.9$$.