SAT Two-Variable Data: The Complete Guide

For exact equally spaced data $12,18,24,30$, consecutive differences are 6, supporting a linear model. For $12,18,27,40.5$, consecutive ratios are 1.5, supporting an exponential model. The same starting values can therefore lead to different models. With noisy data, look for an approximate pattern rather than demanding exact equalities.

In $y=80(1.06)^t$, the initial value is 80 and the growth rate is 6% per unit of t. In $y=80(0.94)^t$, the decay rate is 6%. An exponent of $t/3$ would make the factor apply every three t-units instead. Read the exponent and the input definition together before naming the growth period.

A line of best fit follows the overall linear pattern rather than connecting consecutive points. A reasonable line has points distributed around it without a clear curved residual pattern. Do not choose its slope using any two raw observations unless the prompt specifically asks for that secant slope. To estimate the slope of a drawn line, choose two readable locations on the line itself.

The intercept may be a mathematical prediction with limited practical meaning. A model built from adult trees may have an intercept at zero trunk diameter, even though zero diameter lies outside the measured range. Interpret the algebra correctly while avoiding a claim that the intercept necessarily describes a real observed tree.

A residual is vertical, measured in response-variable units. It is not the horizontal distance from a point to the line, and it is not the perpendicular geometric distance. A point above the line has observed y larger than predicted y, so its residual is positive. A point below the line has a negative residual.

Desmos can help evaluate an equation at a chosen input or fit a model when exact data pairs are provided. Label the variables before entering a table, and compare the resulting model with the prompt's units and task. A regression output is not a substitute for interpreting the slope, and a fit to observational data does not establish causation.

When estimating a line's slope, a steep visual angle alone is unreliable because axis scales differ. Use vertical change divided by horizontal change in the actual labeled units. A short horizontal range can make a modest numerical slope look steep, while a large vertical scale can make a large slope appear shallow.

A fitted line is not an instruction to connect all the points. Connecting observations in arbitrary order can suggest a zigzag pattern that the model is designed to summarize. Likewise, the best-fit line need not pass through the first or last observation. Focus on overall agreement and the question's stated model.

Suppose $y=12+0.4x$ models a quantity with x measured in minutes. If h measures hours, then $x=60h$, so the model becomes $y=12+24h$. The intercept remains 12 because zero minutes and zero hours represent the same starting point. The slope changes because a one-hour increase contains sixty one-minute increases.

When only the predicted change is requested from a linear model, multiply the slope by the input change. The intercept cancels when subtracting two predictions. This can save time, but it applies to linear models with constant additive change. For an exponential model, the multiplicative factor over an interval depends on the interval length instead.

A residual pattern can reveal a model's weakness. If residuals are positive at low and high inputs but negative in the middle, a straight line is missing curvature. Random-looking small deviations around zero are more compatible with a useful linear fit. You need not compute a formal test to recognize a systematic pattern described in the question.

If a model predicts $y=25(1.2)^x$, increasing x by 2 multiplies the prediction by $1.2^2=1.44$. That is a 44% increase over the earlier prediction, not a 40% increase. Repeated growth compounds. The absolute increase depends on the earlier value, while the relative factor depends only on the input change. This distinction can answer a comparison question without evaluating two large model outputs separately. It also explains why an exponential curve's vertical steps grow even though its percentage rate remains constant.

Extrapolation is a model-based estimate, not automatically an invalid calculation. If the prompt asks what the model predicts beyond the observed range, evaluate it. If it asks whether that prediction is well supported by the observed data, explain the additional uncertainty. Distinguish performing the model's arithmetic from judging the model's reliability.

Use about 20 minutes. In the first five, read several equations and describe each slope, intercept, or exponential factor in words with units. Include one input measured in hours and one output measured in thousands. Do this without computing a predicted value so the interpretation receives your full attention.

Next, solve Solar two-variable-data questions for ten minutes. Include a scatterplot prediction, a residual, and a choice between linear and exponential patterns. For each prediction, label the input and output. For each residual, write observed minus predicted before substituting. Record whether a wrong answer came from arithmetic, a unit error, or confusing the model with a data point.

Finish by explaining one limitation of a model in context. Return a day later for a mixed set with percentages, since exponential factors often hide percentage changes. Once the setup is dependable, use calculator evaluation to reduce arithmetic time while preserving your interpretation checks.