Learn how to find out more about your data in Prism, including quartiles, median, SD, SEM, confidence interval, coefficient of variation, geometric mean, and more. In addition, this video will cover how to test for normality in your experimental data.
You will learn how to:
Learn how to find out more about your data in Prism, including quartiles, median, SD, SEM, confidence interval, coefficient of variation, geometric mean, and more. In addition, this video will cover how to test for normality in your experimental data.
You will learn how to:
This video is part of the Essential Statistics series, presented by Dr James Clark, from the School of Cardiovascular Medicine and Sciences at King’s College London.
Transcript:
Hello, my name is James Clark from King's College London. In this brief walkthrough, I'm going to look at ways of describing your data sets and also how to test for normality within your experimental data using GraphPad Prism. On the screen, you can see the results of four experimental outcomes. You can see data for groups A, B, C, and D which have different end numbers in each group and clearly a different numerical outcome in terms of the data set. We can view this data as a graph, and I have prepared the same data as a scatter graph so we can observe the distribution of our data within each group. You can clearly see that groups A, B and C appear to have a fairly wide distribution of data and appear to represent the outcome of three different experiments. Group D, however, has a very narrow distribution of data, very small error bars, and I would be concerned looking at this data, whether they might be normally distributed.
However, before testing for normality, we're going to ask a few questions of these data and ask Prism to summarize these data in the form of a descriptive statistics. You can carry out descriptive statistics on any column of data in Prism by clicking on the analyze button in either the graph or the table view or click on new analysis within the results window. In this example, I'm going to click on the analyze button from the group data shown here in the graph. When you click on the analyzed data, you get the analyzed data option window and from here you can choose a number of analysis from the selection on the left hand side.
I want to look at the descriptive statistics of these data, so I'm going to click on descriptive statistics from the column analysis option. I don't have to analyze or show the descriptive statistics from all of my data sets, but in this example I'm going to, so I make sure that in the analyze which datasets option box, I have selected A, B, C and D. If you have more columns representing, for instance, different experiments that you don't want to show these statistics for, you can deselect them simply by clicking on the little boxes next to each of the column headers. Having selected your data set, you click on the OK button.
The next window that appears are the parameters for your descriptive statistics. Prism offers a wide array of descriptive statistics that you can look at. By default, it highlights the mean standard deviation and standard error of the mean as well as the minimum, maximum and range of your data. In addition to these basic options, you have a number of other options including the sum of your columns, the quartiles. In the advanced and confidence interval option, you can look at coefficient of variation, geometric means, harmonic means, quadratic means. In the confidence intervals, you can look at confidence levels set at whatever level you want to set them at. Once you've selected the statistics you wish to view, you can click on the OK button and when you do this, the descriptive statistics results window appears on the screen.
On the left hand side you'll see we now have a new results pane called descriptive statistics of group data, group data being the name of the table and the graph that are analyzing. The descriptive statistics window is self explanatory. You can see divided A, B, C, and D, the descriptive statistics that we requested. You can see our end number in the top row and then our minimum percentiles, medium, maximum, range, et cetera, down each column. It's easy to see the results of your descriptive statistics for creating tables or other figures outside of Prism.
Now that we know a little bit about our data in terms of the mean values and distribution around those means, it's good to know whether our data fit to a normal or bell shaped distribution curve. When carrying out a parametric analysis of your data such as a T test or an ANOVA, it's important to know whether your data are indeed normally distributed before you do these tests. We're going to return to our group data table and we're then going to click again on the analyze button. Once more, the analyzed data window appears and from the left hand side we choose normality and log normality tests.
On the right hand side, we can choose which data sets we wish to analyze and on this occasion once more, we want to select all of our data sets. A useful function is the deselect all and select all buttons. If you click on deselect all, all your data are deselected and then you can choose for instance the data sets you wish to analyze in particular or you can click on the select all button and all of the data sets on that given table will be selected. Once you've determined which data you wish to compare, you then click on OK.
The next window is the parameters window in order to carry out normality tests. In this example, I'm going to carry out a normal or Gaussian bell-shaped curve distribution test, but I could also choose a lognormal distribution. There are a number of options in Prism of which method you wish to test your distributions. We actually recommend the D'Agostino Pearson normality test. The one that Prism uses is the omnibus K2 test. The test is very simple and works out how far the distribution is from Gaussian in terms of asymmetry in shape, and then it calculates how far each of your values differ from the value expected with a Gaussian distribution. It produces a single P value from the some of these discrepancies which it then reports. If your P value is less than 0.05, if you select your significance level, for instance here in the Alpha to 0.05, your data will not be normally distributed.
We have an option to create a QQ plot and we also have some other options in how to display our P value as we do in other statistical tests. It is worth noting a little question mark down in the bottom left, of all of these options screens will bring up the very intuitive Prism help screen in order to understand further what these tests will be doing to your data. However, I want to do a normal distribution test using the standard D'Agostino Pearson normality test. So I select this one from our list and I continue with the tests by pressing the OK button. The window will appear in the results section sharing a normality and lognormality test results. We can see on the screen we have data from groups A, B, C and D, The K2 value and the P value is reported and then the significance is given to tell us whether it passed the normality test or not.
It's interesting to see here that group A has failed normality test. You can see that Group B has passed the normality test. Groups C, the data set are too small. We only have an N five here and this was not able to do a full normality test on it. If you remember from the graph, I questioned the distribution of Group D thinking that they were a little bit too tightly bunched to be normally distributed and you can see here it failed the normality test with a P value of two stars. This really shows that these data are not normally distributed and would be unwise to carry out a parametric test on these data.
In general, we can find out some very interesting information about our data sets without doing any statistical tests just simply by using the descriptive statistics to show means, standard deviations and confidence intervals and using a normality and lognormality test to check your data or normal distribution prior to doing any statistical analysis of your data.