A least-squares line is fitted, not guessed, and it arrives in three moves. First the slope: $b = r\,\dfrac{s_y}{s_x}$ — b equals r, times s-y over s-x. The subscripts name the variable, not an order: $s_y$ is the standard deviation of $y$, the response you are predicting, and $s_x$ that of $x$, the predictor you are predicting from. $r$ is the correlation coefficient, a naked number between −1 and 1, and $b$ is the slope in response-units per predictor-unit. Read the ratio as a converter: it turns one unit of $x$ into the matching amount of $y$, and $r$ discounts it for the fit being imperfect.

Second the intercept: $a = \bar{y} - b\bar{x}$ — a equals y-bar minus b x-bar, where $\bar{x}$ and $\bar{y}$ (say x-bar, y-bar) are the two means, and $a$ is the line's height at $x = 0$. This one is free, and here is why: every least-squares line passes through $(\bar{x}, \bar{y})$, the point of averages. Stand at that point and walk back down the slope to $x = 0$; where you land is $a$. Third, the prediction: $\hat{y} = a + bx$ — y-hat equals a plus b x, where the hat means predicted, not observed, and that hat matters.

One warning to carry out of this lesson. A fitted line is only evidence about the range of $x$ the data actually covered. Reading it far outside that range is extrapolation, and the line will answer with total confidence and no justification whatsoever — it does not know the data stopped. Fertilizer improves yield right up until it doesn't, and the line will never mention it.