12.7: Exercise- Exporting the Data for Statistical Analysis

A general design feature of ERPLAB is that we do not have any built-in statistical analyses. If we included a few simple but easy-to-use statistical functions, many people would use those functions whether or not they were appropriate. The alternative would be to provide a huge suite of statistical functions, but then it would be complicated to use, and we would be duplicating the functions available in packages like R. Thus, our approach is to make it easy for people to export the data in formats that can be read by general-purpose statistical packages. This isn’t as easy for you as built-in statistics would be, but I think it’s actually better if the statistical analysis isn’t too easy. Appropriate statistical analysis takes some real thought!

In this exercise, we’ll see how the decoding accuracy values can be exported from eStudio for statistical analysis. Make sure you still have the 10 MVPCsets loaded from the previous exercise (1-5 for the ID decoding, 6-10 for the emotional expression decoding). Select all 10 of them in the MVPCsets panel (but don't include the grand averages) and click Export. You should see a window like the one shown in Screenshot 12.9.

Change the name in the Save Output as field to Decoding_Accuracy.txt. This file will be saved in Matlab’s current folder. If you want to save it elsewhere, you can type in the full path or click the Browse button. MVPC value to export should be set to proportion correct, which is the only available option for these MVPCsets. Later, I’ll explain how you can instead quantify decoding accuracy as the area under the receiver operating characteristic curve (AUC).

In the Transpose Data section, you can either have one row for each participant and a separate column for each time point, or one row for each time point and a separate column for each participant. For this exercise, select Each column is a time point.

Make sure that the box is checked for Include time values with units of milliseconds so that the output file contains the timing information. And make sure the Precision is set to 3 (meaning that the accuracy values will have 3 digits to the right of the decimal point). Click Export.

You will now have a file named Decoding_Accuracy.txt in the Chapter_12 folder. You can see it by double-clicking on it in Matlab’s Current Folder pane. Alternatively, you can import it into a spreadsheet or a statistical analysis program.

You’ll see that the top row of the file indicates the latency in milliseconds for each time point (from -200 to +780 ms). And you’ll see that the left column indicates the name of the MVPCset (which contains the participant number and whether ID or emotional expression was being decoded). Each of the other cells of the file contains the proportion correct for a given combination of latency and MVPCset.

In general, you can analyze these data in the same way as regular ERP data. For example, you can compute the mean decoding accuracy across the points in a given time range (like mean amplitude in conventional ERP analyses) and compare these values against chance with a one-sample t test (to see if the decoding was above chance). You can also compare these values across groups or conditions with an F or t test (to see if the amount of decodable information differed across groups or conditions). Alternatively, you can use the mass univariate approach in which you perform statistical comparisons at each individual time point and then apply a correction for multiple comparisons. For more information, see Chapter 11 in Luck (in preparation).

---