We can touch upon only a very small subset of data analysis tools and types of graphs in this brief lecture. You may need to use a specific tool for your project; please discuss this with your advisor and research mentor!
Language note: the word “data” is plural. For example, it is wrong to say (and write) “this data shows that…”. You should say “these data show that…”. If you want to refer to a collection of data, you can say “this dataset shows that…”.
Organizing and Understanding Your Data
- Types of measurements:
- Nominal or categorial, e.g., ‘Male’, ‘Female’.
- Ordinal or ranked, e.g., ‘Low’, ‘Medium’, ‘High’.
- Interval or continuous, e.g., 2.9, 4.2, 3.8.
Visit http://web.uccs.edu/lbecker/SPSS/scalemeas.htm for a good primer on these concepts. They affect what kinds of statistical analysis you will use.
- Types of variable:
- Dependent: the variable(s) you are interested in being able to explain and/or predict.
- Independent or predictor: the variable(s) you believe might affect the value of your dependent variable(s).
A statistical model is a particular combination of predictor variables that you create as a candidate to explain the variation in one or more dependent variables. Statistical analysis is really about choosing the ‘best’ explanatory model.
Recording, Summarizing, and Curating Your Data
The data you have gathered (or will gather) for your thesis is likely to come in many different forms depending on your individual project. However, there are some guidelines that are broadly applicable to everyone.
At some point you are likely to enter your data from its original format (field forms, lab notebook, etc.) into a computer spreadsheet. Think carefully about the best way to arrange your data and keep the raw data together in a master file. Give this file a clear an obvious name. You will likely make several modifications to this file (which you should save as different files) but you should always have an unaltered master file with just the raw data.
As you work with your data, you will manipulate it in several ways. You will almost certainly make summaries (calculating means, frequencies, etc.), examine subsets of data, and manipulate the data to accommodate particular kinds of analyses. You will make your life remarkably easier if you keep a log of your manipulations and analyses. If you spend an hour working with your data, take a few minutes to journal what you have done (either in a notebook on in a dedicated computer file). It is important to record what you did, why you did it, what the results were, and what new files you generated as result. Since you will spend a significant amount of time with your data over the course of writing your thesis, taking a few minutes to record your incremental progress will save you time down the line.
Some specific points:
- Keep your data chains short. As much as possible, derive any subset datasets directly from your master file, so that any changes to the master have to propagate only one step.
- Use live links for your summaries. Excel (for example) allows you to link cells together with formulas. Often, you can generate the summaries, subsets and graphs you want using formulas that link to the master dataset. Then, if you change the mater, everything else changes automatically.
- Give your files meaningful names. Saving files as thesis.xls and summary.xls will not tell you at a glance what a particular file is about. Use a few keywords in your file name to keep data organized (e.g. SiteA_logtransform.xls).
- Metadata. Metadata are ‘data about data’; information on how the data were acquired, what the codes mean, etc. You should record all of this metadata on a separate worksheet associated with the master file. As you make new datasets, make sure you take the time to update the associated metadata. For example, if you log transform you data and save it as a new file, be sure to update your metadata to include any new column heading. You may want to keep the relevant parts of your analysis log on the same page as the metadata.
Reporting data: Significant digits
Be careful to not report more significant digits than is justified by the precision of your data. Typically, the last significant digit is uncertain. There are rules how to determine the number of significant digits to be reported. For a detailed discussion, see this website.
Frequently used basic statistical analyses
- Basic summary statistics, min, max, average…..
- Please report your data with uncertainties (standard deviation, standard error, confidence inteval)
- Simple linear regression can be used to determine a straight-line equation describing the average linear relationship between two variables. MS Excel makes this very easy by allowing you to plot the data and then use the trendline command to fit a line to the data. However, the trendline equation or the R2 does not tell you if indeed there is a significant correlation between the two parameters. For an answer to this question you’ll need to perform a statistical analysis using the regression tool in the Tools/Data Analysis menu. If this option does not appear in your Tools menu, you need to install the analysis toolpak & toolpak VBA under Addins.
A special case of a linear regression is a calibration curve
- A ttest is used to compare the mean of two populations. Again, ttests can be found under the Tools/Data Analysis menu. There are paired and unpaired ttests.
- Chi-square tests – Matt? (if necessary)
- ANOVAs – Matt
Types of plots (This is not an exhaustive list)
- Y versus X scatter plot: simple line or symbol plot.
- Time series: data points are plotted versus time
- Linear regression: scatter plot plus best fitting trend line
- Moving average: data are averaged in blocks around a central point, (for example 10 points on either side of a given point). Makes the most sense for time series data with large amounts of variability. Problems: data points at the very end and start of the time series cannot be included.
- Residual plot: (a fitted trend line, e.g., from a linear regression, is subtracted out, so that only the differences between the trend and the data are plotted).
- Pie chart: good for displaying data that should add up to 100%.
- 3D plots: if you have measurements that depend on two other variables, you could plot the measurements as ‘heights’ (z values) on a ‘landscape’ of the variables (x and y values).
- Log-log or semi-log plot: used for displaying data with large ranges in the numbers (for example, data points that range from 1 to 1000). Problem: can obscure serious errors in the data.
Characteristics of Good Plots, Figures, and Tables
A key feature of good scientific communication is making good figures and plots. Many more readers will see your figures and plots than will ever read the text of the paper.
- Symbols are legible and distinctive. The symbols are large enough to view and distinguish if the page is held at arms length.
- Lines connecting symbols are legible and distinguishable (where possible).
- Every figure has a figure caption that explains the overall purpose of the figure and the meaning of every symbol and line on the figure, if no legend was included in the figure.
- The plot is not overly busy. Too many symbols and lines on one plot are simply confusing.
- Where appropriate, symbols should have error bars.
- Don’t put titles on graphs, all the information about the graph should appear in the figure caption (Figures in presentations can have titles, they usually don’t have captions).
- A good plot would be legible if it was shrunk down onto a 3 by 5 index card. This is also true for a good table.
- Tables are difficult to display in talks, but are vital for papers. They can compress information and avoid boring, repetitious discourse. They can also help to keep you organized. Tables for talks should be VERY simple.
- Try stacking plots that are related to avoid overly busy, single plots.
Avoid becoming a graphical sinner!
- Do not use numbers or graphs in such a manner that — either by intent, or through ignorance or carelessness — the conclusions are unjustified or incorrect!
- Example: “Soaring overseas stake of Japan’s multinational companies” (from Business Week, June 16, 1980; after Jaffe and Spirer, 1987) (Fig) and redrawn with a consistent scale on the horizontal axis (Fig).
- Example: “Birthrate soars in decade,” Birthrate expressed in per 1000 and year (Fig).
- Depressing monthly sales (Fig) and encouraging cumulative sales (Fig).
- Back up your data.
- There are many software packages out there that allow to make more or less sophisticated plots.
- MS Excel is limited but does cover most of the data analysis/plotting tasks that you need to perform (you may have to add the data analysis tool pack as an “Add in”).
- You can make simple figures (sketches, flow charts, etc…) in MS Word/PowerPoint by using the drawing tool bar at the bottom of the window.
- All kinds of figures created in other programs can be incorporated in MS Word/PowerPoint. Include a figure caption!
- Back up your data.
- If you want to use scanned images, the resolution of your image should be ~200dpi at the size they will be ultimately printed. For PPT presentations a typical projector has a resolution of 1024×768 pixels. Depending on how much area your image is covering, you should scale your image size accordingly. Save photo-type images as JPG files, and diagram-type images (few colors, sharp transitions) as GIF files.
- Use the ‘Paste special’ option and paste images as ‘Picture (enhanced Metafile)’.
- In order to reduce file sizes, unclick the ‘fast save’ option in the Tools/Options menu in MS Word/PPT.
- Find out what the size limit is of your e-mail sevice provider. Sending e-mails > 3MB often is problematic.
- Did we mention… Back up your data!
Where to find help
- Statistical Consulting Services – The Department of Statistics at Columbia University
- Statistical advice at any stage of research (sorry no help on homework problems or class projects).
- You are encouraged to come in during the early stages of your research so consultants can be helpful at the design stage.
- Room 1025 SSW Building
- See website for current walk-in hours or if these times do not work arrange an appointment
- Statistics software packages
- SPSS – available on many lab computers
- Help files/tutorials
- Most software packages have extensive help features and tutorials.
- Microsoft Excel help
- Gotelli, N.J. and A.M. Ellison. (2004) A Primer of Ecological Statistics. Sinauer Associates, Sunderland, MA. 510 pp.
- Jaffe, A.J., and Spirer, H.F. (1987) Misused statistics. Marcel Dekker, Inc., New York, 237pp. (HA29.J29 1987)
- Berk K.N. and Carey, P. (2000) Data Analysis with Microsoft Excel. Duxbury, Pacific Grove, CA, 587pp.
- EESC 3017 Environmental Data Analysis
- Example spreadsheet (graphs.xls)