Many articles have been introduced on the comparison of statistical analysis software, but most of them are aimed at professional statisticians. In medicine, the vast majority of people who use statistical software are actually clinicians who are not majoring in statistics. For various reasons, they had to teach themselves to use statistical software. Based on more than ten years of experience in medical statistical analysis, the author is well aware of the difficulties faced by clinicians in using statistical software on their own.
In order to help more clinicians get rid of this difficulty, help clinicians learn how to use JMP to efficiently carry out data analysis, and improve the efficiency of daily work and publication of papers, from now on, Dr. Feng Guoshuang, a senior JMP user and senior statistician, He and his team will share a series of statistics and data analysis, JMP practical operations and other practical content on the JMP data analysis platform. Each issue has a classic topic to help everyone master a new skill. It is worth noting that these topics are not just for clinicians, but are applicable to everyone who uses JMP software to perform data analysis.
This article is the first in a series of articles. Dr. Feng and his team share their journey and experience in using JMP software, and give examples of why JMP software is a statistical software more suitable for clinicians. To facilitate reading, the article is described in the first person.
So, why do we think JMP software is more suitable for clinicians to use? There are actually many reasons, here are a few main ones:
01 Full-featured table-style operation
This is probably what clinicians are most concerned about. Because statistics is not the main job of clinicians, there is no need to spend a lot of time learning statistical software, and menu-based operations are definitely more in line with the needs of clinicians than program design.
You can intuitively feel the function table of JMP (Figure 1).
Figure 1 JMP analysis menu
From the analysis function table, we can see some features of JMP in statistical classification. It is not simply classified according to specific methods, but a broad classification based on the structure of statistical methods.
For example, if you want to compare the means of two groups, from a statistical point of view, the so-called "comparison between groups" is actually a single factor analysis. The group is the index, and the indicator to be compared (outcome indicator) is the dependent variable. Statistics In mathematics, the argument is referred to as x and the dependent variable is referred to as y. Therefore, comparison between groups means analyzing the impact of group factors (initiator x) on outcome indicators (dependent variable y).
Therefore, in JMP, all between-group comparisons, or more broadly, all one-way analyses, are contained in the Fit Y by X menu.
This approach is helpful for understanding statistical thinking. Some clinicians may not be used to JMP's method at first, but once they use it, they will find that this method is actually more convenient because it is divided into broad categories according to statistical methods, which is more conducive to understanding statistics. Associations between methods.
In addition, the JMP menu can be switched between Chinese and English at any time. For scientific researchers who want to publish English articles, they can first make analysis results in Chinese, and then change the language directly in the menu (Document→Preferences) English, effective immediately, no need to restart. This is very humane.
02 More convenient analysis process
In the past, in most menu-based statistical software, when you want to use a certain statistical method, you click on the corresponding menu, and then there are various options waiting for you. After all settings are completed, click OK to directly enter the results section.
For example, suppose we have a piece of data (partial display) shown in Figure 2, and now want to see whether there is a statistical difference in physical health scores between people who drink and people who do not drink.
Figure 2 Sample data
If you use a certain software to compare these two groups, the analysis function table is as shown in Figure 3. You need to set up various things, then confirm and enter the results interface.
Figure 3 Analysis interface of a certain statistical software
JMP is different. The analysis menu is very simple. There are no options. Just drag and drop the required variables to the corresponding position (Figure 4). No need to make extra settings, JMP will automatically determine which method results should be entered based on the type of x (categorical or continuous, as shown in Figure 4, the red variable is categorical, the blue variable is continuous) and the type of y.
In Figure 4, since x is a categorical variable and y is a continuous variable, the software automatically selects the "single factor" (see the lower left corner of Figure 4, equivalent to variation analysis) method. If x and y are both categorical, the lower left corner of Figure 4 will become "contingency" (equivalent to the chi-square test).
Figure 4 JMP analysis interface
Some people may ask, there is no option, so what if you want to make some changes? For example, although y is a continuous variable, what if it does not satisfy the normal distribution and needs to be tested without a parent number? This is the specialty of JMP. The JMP analysis interface is very simple, but the results interface is very rich, and all options are in the results interface (Figure 5).
When we click OK in the Figure 4 interface, we enter the JMP results interface (Figure 5). The resulting interface is simple by default, but has rich options.
Figure 5 The analysis results interface of the "Fit Y with X" menu in JMP
As can be seen from Figure 5, if you want to display other results, you can check the methods you need through the red drop-down menu (red triangle) at any time. If you no longer need them, you can also check them off at any time. This approach makes data analysis and exploration more convenient. In other software, if you want to do it again, you have to go back and do it all over again. But it is not used in JMP. If you want to change the method (for example, you did a t test and suddenly found that the data may not conform to the normal distribution, and you want to do a motherless test), there is no need to return, just check other methods in the results. ,Very convenient.
03 Semi-intelligent operation
All clinical researchers may have a dream, that is: how great it would be if they could log data into software and then leave everything to the software.
Although there is no fully intelligent statistical software, semi-intelligent software can still do it. In other words, although statistical software cannot choose a method entirely based on the data, it can still help you make a choice in some specific situations. For example, if you want to compare the means of multiple groups, you definitely cannot use the t test directly; if you want to do a regression analysis, and the outcome is categorical data, in most cases, you will first consider logistic regression, etc. This is where JMP helps everyone become semi-intelligent.
In fact, from Figure 4 introduced earlier, I believe you have already felt that in the "Fit Y with
Let’s further demonstrate the performance of JMP in semi-intelligent mode.
Let us illustrate again by comparing the physical health scores of two groups of people who drink and do not drink. Let’s first review the results in Figure 5. Since it is a comparison between two groups, t-test can usually be considered. We note that there is a t-test in the results (“merged t” in Figure 6).
Figure 6 Analysis results interface for comparison between two groups in JMP
Now we change a grouping variable and compare the physical health scores between different cardiac function grades. Cardiac function classification is a multi-classification, and variance analysis is first considered. We use the same operation method as in Figure 4 (just replace drinking in Figure 4 with cardiac function classification) to get the analysis results interface in Figure 7.
At this time, you will be surprised to find that the "merged t" in Figure 7 is missing.
Figure 7 Analysis results interface for multiple group comparisons in JMP
why? Because t-test cannot be used for multiple group comparisons, JMP automatically removes methods that should not be used for you. This is very important, especially for non-statisticians.
For another example, suppose we use the physical health score as the dependent variable to analyze the impact of several other indicators on the physical health score. In general, linear regression can be considered.
We drag the dependent variables and arguments into the corresponding positions through the "Fit Model" menu (Figure 8). Then take a look. JMP selects the "standard least squares method" (equivalent to linear regression) by default, and automatically grays out the two logistic regression methods, leaving you unable to choose.
Figure 8 Analysis interface of the “Fitting Model” menu in JMP
Why is this? Because the outcome (physical health score) is a continuous variable, logistic regression cannot be used for regression of continuous variables, and linear regression is the most commonly used. For this most basic mistake, JMP directly helps you avoid it.
Through the above two examples, you can see that JMP software reflects its love for non-statistics majors in every aspect. Professionalism can be seen in the details, which is really well-intentioned.
04 Fully interactive operation
Most statistical software works from data to results. After making results based on data, it is generally difficult to go from results to data.
But JMP is different. The results are just the beginning of the real analysis. Not only can corresponding results be added or deleted according to needs (Figure 5), but the results are always related to the data.
Figure 9 shows the distribution of age, provides statistical description indicators, and provides graphics. The key thing is that these graphs can be selected with the mouse, and when you select the graph of any variable, the corresponding observations of other variables are also selected.
For example, in Figure 9, there are two variables, age and age stratification. When we select the corresponding data of the variable age, you will find that the variable "age stratification" is also selected, and "age stratification" The corresponding one in the frequency list is also selected. If you look in the data set, you will find that these data are also selected in the data set. In other words, JMP software is not simply "data → results", but data and results are interactive. It can be "data → results" or "results → data". Data and results are always related.
Figure 9 Example of interaction between data and results in JMP
Some people may think, what is the use of this connection? Very useful in practical work. For example, if you want to select "drinking" people separately, you don't even need to use these filtering functions. You can directly select "drinking" in the picture, and then it will be selected simultaneously in the data set (Figure 10), and then use the "subset" function Just select the information of drinking people.
Figure 10 Using graphics to filter data in JMP
For another example, if we find an abnormal point in the scatter plot and want to quickly locate the data, we can directly select the point in the graph, then go to the data set and find that the point is marked, and we can directly use the positioning function to locate it. This observation saves you the time of checking this information yourself.
05 Other humanized functions
JMP actually has many user-friendly functions, which are difficult to finish in this article. They will be introduced one by one in subsequent series of articles. Finally, let me mention a few more practical and time-saving functions.
For example, in JMP data tables, there are no restrictions on variable names and variable value naming. There is no problem whether it is Chinese, English, special symbols, etc. At present, most statistical software cannot do this, and often has certain restrictions on variable names, which results in the need to modify the figures after the results are obtained, which is too troublesome. JMP can directly produce the result based on the variable name without modification.
For another example, the graphing function of JMP is interactive, which is different from other statistical software in the past. It does not rely on specifying variables to draw graphs, and it can be realized entirely by dragging and dropping (Figure 11), and multiple graphs can be freely To switch, just click the mouse (the upper graphic in Figure 11, switch instantly with a click of the mouse).
Figure 11 JMP drawing interface display
Of course, the advantages of JMP are not limited to this. The above are just some experiences based on what we have used in actual work. In the subsequent series of articles, we will also introduce the characteristics of JMP in data sorting and cleaning, graphing, and statistical analysis one by one through cases and explanations. We believe that these will provide everyone with faster and more efficient analysis methods.
