Tables, Graphs and Transformations reviews graph selection, coordinates, intercepts, gradients and biological interpretation for Marine Science practical study. Check the current syllabus and question instructions for the assessed scope. A graph presents data and may also show a fitted model. Its design should preserve variable type, measurement scale and uncertainty.
Graphing study scope
Practise drawing and interpreting graphs using the display appropriate to the variables and purpose. Continuous paired measurements, categories and grouped frequency distributions do not all call for the same graph.
Match axes to the quantities and units in the table. Choose readable scales that use the plotting area effectively, and small symbols whose centres can be located accurately. Choose a fitted line, smooth curve or ruled joins according to the data and instructions; justify any extrapolation.
This note also practises error bars, calibration curves, gradients, tangents and transformations between numerical, graphical and algebraic forms. Confirm level-specific assessment requirements in the current syllabus.
Table structure before graphing
The table is the source for axis labels and plotted values. Keep the independent variable first, raw observations visible and calculated means or rates in separate columns. Distinguish independent units from repeated readings and subsamples.
Use one unit per column and retain justified reading resolution and appropriate calculation rounding. A graph cannot repair rounded or mismatched source data.
Flag unusual results for investigation, preserve the raw values and document any exclusion rule. Being unusual or failing to match a prediction alone is not a reason to omit a value from a mean. If all replicates are plotted, do not conceal inconvenient points.
The recording note owns full table construction; this note uses the table to make a defensible visual model.
Choose the graph from data type
Continuous independent variables such as temperature, salinity, pH, time, depth or distance have meaningful intermediate values. A line or curve may be appropriate when connection or interpolation is scientifically justified; continuity alone does not require joining the measured points.
A bar chart compares separate categories such as habitat type, species or treatment identity. Bars are separated because the categories do not form a continuous numerical scale.
A histogram shows a frequency distribution for continuous measurements grouped into class intervals. Adjacent bars touch. When class widths differ and area is to represent frequency, use frequency density: class frequency divided by class width. For a probability-density histogram, also divide by the total observation count, so the total area is one.
Check this topic from memory
Attempt the matching topic bank before reopening the notes. Use each missed idea to decide what to review next.
A scatter graph examines association between paired variables. It does not by itself establish causation.
Independent and dependent axes
Plot the independent variable on the horizontal x-axis and the dependent variable on the vertical y-axis unless a supplied convention requires otherwise.
Copy descriptive quantity names and units from the table. "Time / min" is useful; "x" or "results" is not.
For a transformed graph, label the transformed quantity explicitly, such as reciprocal time or logarithm of abundance. Do not label it as the original variable.
Category labels belong at bar centres, while histogram boundaries mark interval edges.
Choose a usable scale
Use the available grid effectively while keeping values easy to plot and read at the required resolution. Simple increments such as 1, 2, 5 or their powers of ten are usually effective.
Axes do not always need to begin at zero for line or scatter graphs, but a break or non-zero origin must not mislead. Bar-chart length encodes magnitude, so a zero baseline is normally important.
Mark scale values at regular intervals and check that every point fits before plotting.
Avoid awkward increments that turn every coordinate into mental arithmetic and increase plotting error.
Plot points accurately
Use a sharp pencil and small cross or small dot in a circle. The intersection or centre marks the coordinate.
Plot one series completely, then check values against the table before adding another. Use a clear key or direct labels when several series share axes.
Large blobs obscure the intended coordinate and reduce plotting and reading resolution.
Draw error bars according to the statistic and interval specified. Mean plus or minus one standard error gives symmetric bars around the mean; other intervals may be asymmetric.
Decide how points should be connected
Use ruled straight joins for a sequence where each point represents a defined successive state and interpolation between measured points is intended without assuming a smooth global model.
Use a smooth curve for a biological response expected to change continuously and nonlinearly, such as a rate approaching an optimum or plateau. The curve should follow the central pattern rather than pass through every point.
Use a straight line of best fit when the relationship is approximately linear. For a hand-drawn estimate, follow the overall trend across the range rather than forcing the line through individual points. Ordinary least-squares fitting minimises the sum of squared vertical residuals; it is not defined by equal counts of points above and below.
Do not connect scatter points in observation order when the purpose is correlation.
Best fit and anomalies
A line of best fit represents the overall pattern and can reduce the influence of random scatter. It need not pass through the origin unless theory and data support that constraint.
An anomalous point lies away from the main pattern. Check plotting and transcription first, then method evidence. Do not redraw the line to force inclusion or erase the point.
State whether the anomaly is omitted from a calculation and why. A surprising point can indicate real biology, not merely error.
With very few or widely scattered points, a confident best-fit model may be unjustified.
Error bars
Error bars can represent a range, standard deviation, standard error or a confidence interval. State exactly which statistic or interval is shown and, where relevant, its multiplier or confidence level.
For the same type of uncertainty interval on comparable scales and with appropriate sample designs, shorter bars indicate a more precise mean estimate. Overlap or non-overlap of separate bars is not a universal significance test. Assess a difference using the specified comparison method and assumptions, including whether observations are paired or independent; do not infer a p-value from overlap alone.
Standard deviation describes spread among observations; standard error describes the sampling variability of an estimator, here the mean. Calculating it requires the appropriate independent-unit and sampling assumptions. Bars must not be interpreted as the full range unless they represent it.
Unequal sample sizes can affect standard error and comparison.
Describe a graph before explaining it
Identify direction, shape, key ranges, maximum or minimum, plateau, threshold and anomalies. Support the description with coordinates or calculated changes.
For example: the measured oxygen-change rate increased steeply from the first to third light value, then increased more slowly and approached a plateau. If the method measures net dissolved-oxygen change, do not relabel it as gross photosynthesis.
Explanation adds mechanism, such as light ceasing to be the limiting factor. Keep description and explanation distinct.
Avoid writing only "positive correlation" when the graph shows a curve, optimum or changing gradient.
Interpolation and calibration curves
Interpolation estimates a value within the measured range. Draw from the known coordinate to the best-fit line or curve and then to the other axis, reading with the graph scale.
A calibration curve links known standards to measured response. An unknown response can be converted to concentration only within the validated range and under the same method conditions.
If the unknown lies beyond the standards, dilute it into range or prepare a wider calibration rather than extending the line casually.
Report only the digits justified by the graph scale and the uncertainty of the fitted relationship, not extra calculator digits.
Extrapolation
Extrapolation predicts outside the measured range. It is less secure because the relationship may change, reach a limit or become biologically impossible.
State the assumption that the observed relationship continues and explain the evidence and limits supporting that extension.
Avoid extending curves into negative abundance, impossible percentage cover or lethal conditions without evidence.
Whenever possible, collect additional values instead of relying on distant extrapolation.
Gradient of a straight line
Gradient is change in y divided by change in x:
gradient=ΔxΔy
Choose two well-separated points on the best-fit line, not necessarily raw data points. Draw a large gradient triangle and include units from y divided by x.
The sign shows direction. Magnitude shows rate of response per unit independent variable.
Do not invert the ratio merely because the x values are larger.
Tangent to a curve
Where a curve has a finite derivative, its tangent passes through the point of interest with that derivative as its gradient. A tangent can cross the curve, so touching without crossing is not the defining rule. Draw the longest reasonable tangent that does not simply follow a wide section of curve.
Calculate its gradient using two well-separated points on the tangent. This estimates instantaneous rate of change at that coordinate.
A chord between neighbouring data points gives an average rate, not the tangent rate.
On a plateau, the tangent gradient approaches zero; at an optimum it can change sign.
Area under or between graphical regions
When a question defines biological meaning for area, estimate it with geometric shapes or a stated numerical method. Keep axis units, because area units are their product.
Do not assume every graph's area has scientific meaning. A concentration-time area and an abundance-distance area require context.
For irregular regions, smaller subdivisions improve approximation but should not create false precision.
Show the component calculations so the reasoning can be checked.
Transformations
A transformation changes variables to reveal a relationship, stabilise scale or support a supplied model. Examples include rate from reciprocal time, percentage change, logarithm, square root or normalisation per unit area or mass.
Transform every value consistently and show a sample calculation. A reciprocal is undefined at zero but is defined for negative non-zero values. Real logarithms require positive arguments; real square roots require non-negative arguments. Check both mathematical domain and scientific meaning before transforming.
A straight transformed plot does not erase biological assumptions. Interpret the transformed gradient and intercept in the context of the given relationship.
Do not transform only the points that fail to fit.
Reciprocal time as a rate proxy
In a clock investigation, reciprocal time can serve as a relative rate proxy when the endpoint represents a comparable amount of change under each condition and the measurement method remains comparable:
relative rate=time1
Its unit is reciprocal time. It is not an absolute production rate unless the fixed amount of change is known.
Transform individual times or justify the chosen order of averaging because reciprocal of the mean is not generally the mean of reciprocals.
Very small times amplify timing uncertainty.
Logarithmic and percentage displays
Log scales can display values spanning several orders of magnitude and can convert some multiplicative relationships into linear patterns. Axis labels must make the base and transformed quantity clear.
Percentage change compares change with the starting value. The usual calculation is undefined when the initial value is zero and can be highly sensitive to measurement uncertainty when it is very small.
Normalising abundance per unit area or sampling effort enables fair comparison only if detection and method remain comparable.
A visual compression of differences on a log axis should not be described as a smaller absolute effect.
Multiple series and keys
Plot multiple treatments on the same axes when direct comparison is valuable and units and scales match. Use distinct permitted symbols or line styles and an unambiguous key.
Do not use colour as the only distinction in an examination drawing. Avoid clutter that hides individual points and error bars.
If scales differ fundamentally, separate panels may be clearer than a misleading second axis.
Compare series at the same x values and consider uncertainty before claiming one is greater.
Worked application: oxygen production against light
In this illustrative design, mean net dissolved-oxygen change per unit time is measured at six light-intensity values, with standard errors calculated from appropriate independent units. Plot light intensity on the x-axis and mean rate on the y-axis, copying units from the table. Choose simple scales that use most of the grid, plot small crosses and add vertical standard-error bars. A smooth curve is appropriate if rate rises then plateaus; do not join scatter mechanically. Describe the steep rise and plateau with coordinates before explaining that another factor becomes limiting. Estimate the rate at an intermediate measured-range intensity by interpolation. To find local sensitivity, draw a tangent at the requested intensity and calculate change in rate divided by change in light using a large triangle. Extrapolation beyond the highest light value is weak because heating, photoinhibition or a new limit may alter the relationship.
Common misconceptions and corrections
Choosing every graph as a line graph. Data type determines display.
Using joined bars for categories. Bar-chart categories are separated.
Leaving gaps in a histogram. Continuous classes are adjacent.
Using raw frequency heights for unequal bins when area represents frequency. Use frequency divided by class width.
Putting the dependent variable on the x-axis. The independent variable normally belongs there.
Labelling axes only x and y. Use quantity and unit.
Keeping an original label after transformation. Name the transformed variable.
Using only a small corner of the grid. Scale should use most of it.
Choosing awkward scale increments. They increase plotting error.
Starting a bar-chart axis above zero without warning. Bar lengths become misleading.
Plotting large blobs. The coordinate becomes unclear.
Omitting a key for multiple series. Data identity is lost.
Joining correlation points in order. Use a scatter pattern and best fit.
Forcing a best-fit line through every point. It represents the central trend.
Forcing the line through the origin. Theory and evidence must justify it.
Deleting an anomalous point from the graph. Preserve and evaluate it.
Calling every curved response a positive correlation only. Describe shape and range.
Explaining before describing. Establish the evidence first.
Calling standard-error bars the data range. They describe mean precision.
Treating error-bar overlap as a significance test. Use the stated interval and the appropriate comparison method.
Interpolating outside the data range. That is extrapolation.
Extending a calibration line far beyond standards. The response may change.
Reporting more precision than the graph supports. Read to the scale.
Using raw data points for a best-fit gradient. Use points on the line.
Using a tiny gradient triangle. Reading uncertainty becomes larger.
Calculating change in x divided by change in y. Gradient is change in y over change in x.
Giving gradient without units. Axis units determine them.
Using a chord as a tangent. It gives average rather than local rate.
Assuming every graph area has meaning. Context must define it.
Transforming selected inconvenient points only. Apply one rule to all values.
Taking a reciprocal of zero. It is undefined.
Calling reciprocal time an absolute rate automatically. It may be only a proxy.
Using logarithms without naming the base. The scale is ambiguous.
Claiming a log graph reduces the absolute difference. It changes display scale.
Using a second axis that hides incomparable units. Prefer clear separate panels.
Assessment guidance
Identify data type before choosing a line graph, bar chart or histogram. Copy quantity and unit labels from the table, select simple scales that fill the grid and plot precise small symbols. Choose ruled joins, a smooth curve or best fit from the scientific meaning of the points. Preserve anomalies and state what error bars represent. Describe direction, shape, range and key coordinates before explaining them. For interpolation, remain within the measured range; for extrapolation, state the continuation assumption. Use widely separated points on a best-fit line or tangent, show change in y over change in x and include units. Label every transformation and interpret it in the original biological context.
Retrieval practice
Classify ten marine datasets as line, bar, histogram or scatter displays and justify each. For one dataset, design axes and scales, plot two series with error bars and select the correct line treatment. Then read an interpolated value, evaluate an extrapolation, calculate a straight-line gradient and a tangent gradient, and transform a fixed-endpoint time series into reciprocal-time values with correct units and limitations.
