JMP is Not a Spreadsheet

[#1]

It seems to me that a lot of requests appear in these discussions because users attempt to solve a problem or use JMP like a spreadsheet. The rows and columns in a JMP data table resemble a spreadsheet but this appearance is misleading. My intention here is to discuss ways of using JMP, interactively or scripted, that will be rewarding instead of frustrating. I hope that others will join this discussion. Maybe we can reduce the level of frustration in the future that arises from attempts to use JMP in a way that it was not intended.

First of all, a spreadsheet is oriented around a cell. A cell is a location in which to store a value or a formula. It accepts formats. It can be organized with other cells in rows and columns. It is easy to work with rows, columns, or a selection of cells, but they are still individual cells. I can put anything anywhere at any time. That behavior is convenient when working in software that is optimized for consolidating and reporting financial data. JMP, on the other hand, is software for discovery about and between variables using myriad statistical analyses and visualizations. It is oriented around the variable. A data column is a cohesive collection of values for each variable. These related values share the same meaning and, therefore, the same format and other meta-data. A row is also a cohesive collection of values that represent an observation and share row states.

Second, the data table is primarily for storing and organizing data (variables and observations) along with their meta-data. It is not responsible for any kind of analysis. The numerical and graphical analyses happen in the many specialized platforms available through the Analyze and Graph menus. The platforms work with the data table and data filters (change row states). Multiple platforms may be simultaneously opened on the same data set. Multiple platforms maybe combined into a single window when this enhances the analysis.

How else is JMP not a spreadsheet? I will be back with more ideas but it is now your turn.

No software can claim to be all things to all users. I would not expect a word processor to be good at functional data analysis nor would I expect it to be easy to teach it to do so. Many different kinds of software easily and successfully work and play together today so that we may use each of them to their best advantage.

[#2]

I do not know if I can add much more to what has already been stated here, but here goes.

At my insistence, my employer purchased JMP so that I could perform the type(s) of analysis required for and by our most demanding client. While most government contracts are gray, this particular client wanted to go deeper than “what would happen if we implemented this?”¹ Or, as he put it, he wanted his study “to have a methodology and results that were defensible.” That meant following a statistical process: design of experiments, a data collection plan based on the DoE, modeling and analysis of the data, and inferencing the results from statistical analysis.

Prior to our acquisition of JMP, everything was performed in Excel. Depending on who was processing the data the workbooks that were created ran anywhere from mediocre to why. Despite the large user base, Excel is still at its core a powerful accounting ledger, so what our management incorrectly deemed to be statistical analysis was nothing more than summary analysis.² Proving this particular client’s need was not going to happen in Excel.

When we acquired a JMP license, my first thought was that a data table looks very similar to a spreadsheet, but I quickly found and realized that where Excel and other spreadsheet software typically have the cell as the base object; that is, each of the 6,871,947,674 cells is independent having no relation to any other cell unless otherwise defined by the user. As I had prior experience in SAS, it became rapidly apparent that a data table is a data model in waiting. As others have noted, columns serve as either independent variables (predictors) or dependent variables (responses), while the rows in a JMP data table are observations.

JMP’s paradigm prevents users from engaging in some of the more atrocious way in which people use Excel; such as putting multiple “tables” on a single worksheet instead of being treated as independent objects that should occupy their own worksheet. Excel while seemingly having a broad range of graphs, pales in comparison to the Graph Builder platform.

The various platforms offered in JMP have no equivalent in Excel outside of perhaps third-party add-ins. For people inexperienced with statistics, the Analysis ToolPak add-in that included, but not pre-installed in Excel reinforces the misconception that there is not much to performing statistics. Want to design an experiment? Can you employ various techniques to determine if a data set is from a normally distributed population? Need to perform analysis on non-parametric data? Do you want to perform pairwise analysis to determine the factors in your model that are significantly different? If you are using Excel, forget about addressing such needs.

Recently, a co-worker an I showed the president of the company how he could quickly explore data in JMP. The Distribution platform alone sold him, as he saw that with little effort he could rapidly see how data was distributed with histograms and box plots, as well as get quantiles and summary statistics. The capabilities of Graph Builder blew his mind. Our president later asked me if JMP can perform SQL queries, t which I answered, “Yes.” I provided him with the built-in PDF books included with JMP where he how to perform such queries.

Simply put, true data discovery and statistical analtysis software, such as JMP, are far better suited to analytics than the kludges that are applied in Excel. Excel is not designed to perform much beyond reporting summary statistics and basic graphs.

¹ It is often the case that government contracts are posed by non-technical persons at the management level. More often than not, the contractor needs to guide their client to a solid, and achievable, objective given the biding price and contract period of performance.
² One of the major issues data analysts and data scientists face is getting their supervisors and clients to realize that the generation of descriptive (summary) statistics is not the application of statistics, or more properly, statical science and data analytics that leads to statistical inference.

[#3]

I also stumbled on to this thread and thank the previous responders for some great ideas.  I offer my humble experiences.  I do two things in my professional life:

1. I teach experimental design and sampling as means to gather data (and regression when there is no data collection plan) and a host of procedures to analyze the data.  I follow my own guidance in the analysis of data: Practical, Graphical and lastly quantitative.

2. I analyze data sets form clients (these can take many forms).

I make a point that ~95% of my data analysis time is spent organizing the data into a structure/order so the data can be analyzed correctly.  All questions that can be answered, conclusions that can be made, confidence in extrapolating the results, etc. are DEPENDENT on how the data was acquired (context).

All of my clients use Excel.  I gave up on some emphatic DON"T USE THAT POS (piece of software). Software is a language.  Many individuals speak Excel fluently and they are comfortable with it.  JMP is a different language.  Learning a new language can be terrifying and frustrating. Excel has its good points and its issues.  Its good points include how extremely flexible it is, that is, of course, one of its bad points.    I state ALL statistical analysis software requires discipline in its use (yes this includes dare I say Minitab...yikes).  Over the course of working with my clients, it becomes clear the incredible capability of JMP.  I demonstrate how organizing the data wrong leads to erroneous analysis. They naturally become convinced it is in their best interest to recognize how JMP performs analysis.  I have not found one "event" to be the convincing argument, but repetition. Once they see this (admittedly it takes many reps), they now can use Excel with the knowledge of how it will need to be organized in JMP to perform the correct analysis.

Cheers,

Bill

[#4]

A recent JMP Discussions reply from @Mark_Bailey regarding the KS test interpretation illustrates another differentiating global feature in JMP compared to your garden variety spreadsheet application. Interactive/online Help. If one is curious about a specific analysis platform report element and want to learn more, you don't have to hunt/peck/guess by keyword or table of contents search to learn more. All you need to do is go to the JMP main menu bar, select Tools -> Help, then hover your cursor over the element you want to learn more about and click on the element. You'll go directly to the most relevant content in the JMP documentation. Easy peasy.

[#5]

To me, before we address technical differences (such as cells vs. variables), we have to keep in mind that there are "billions-and-billions" of spreadsheets out there created by "millions-and-millions" of long time users.

That is, Excel is an engrained Culture*, while JMP is ... ?

Any thoughts or experiences on how we address the Excel culture when promoting JMP?

* that comes automatically with Office software suites.

[#6]

You have to demonstrate the value-add from a business perspective.  The value is in time-saved in doing the analytics, not only doing the analytics faster, but far more powerfully (and easily).  Start with graph builder.  This platform alone is so incredibly powerful and really becomes easy to use after some patience and fortitude - it's drag and drop.  JMP.com/learn is a great place to start.  Short 5 minute videos are clear and precise to get the user base going.  Seeing is beleiving. Visual analytics are the beginning and the end of every good data analysis.  

Excel doesn't have this functionality.  You can become a "pivot-table" expert with a few short tutorials in JMP in less than 10 minutes.  You can become a "graphing" expert with just a few short tutorials in JMP in less than 10 minutes.  With Excel it's much harder to get there, especially for the average user.  And often the average user has the most direct insight into problems (from the hands-on perspective).

[#7]

Maybe one more way to help facilitate JMP adoption is to install the JMP add-in in MS Excel. It talks to JMP and has analysis icons from JMP within Excel, so that helps for people who really can't relinquish Excel. Excel is a wonderful tool as well and can complement and even supplement certain analyses in JMP where certain calculation options, for example, are not available. One simple example would be a test for equivalence under the assumption of unequal variance (which to my knowledge is not available in JMP). You can build an Excel spreadsheet for this spreadsheet in the special case of for example: if you were to find evidence to reject the assumption of unequal variance using the hypothesis tests for equal variance output.

[#8]

If your talking about test for equivalence of means allowing unqual variances, I believe the can be done, but "indirectly".  Equivalence test of means (as, say 95% level) is given by (90%) confidence interval for difference in means.  Believe the t-test in the Fit Y by X platform gives that confidence interval allowing unequal variances,

[#9]

Thanks @Paul_J, you are right in that I could pull the CI from the 2 sample t-test assuming unequal variances within the Fit Y by X Platform and Compare it to zero to assess the statistical significance (or lack thereof).

What I'm actually talking about is the Equivalence test using TOST (two one-sided t-test approach) where you have to specify a practical difference threshold that defines the maximum difference on average, allowable between groups to be considered not practically important (and by default, that difference is symmetric about zero). For that particular method, there is no option for "unequal variances": the default is equal variance assumed.

[#10]

Indeed, I've had quite a bit of success with the Graph Builder, ... as well as Control Chart Builder and Local Data Filter.  Many Excel users love the "pivot table feel" with power of graphics.

Another sucess I've had at showning value is replacing hours of manal copy-paste and cell formulating in Excel with seconds via a JMP script.

The Help and Books are great also!

[#11]

@Paul_J, this is a key observation. The table manipulation and automation makes tasks that are ridiculously complex or difficult in Excel very fast, easy and reproducible. It opens the door to getting to the analysis you wanted to do, but were too exhausted to even start before.
Another @Patrick GIULIANO mentioned indirectly is that in Excel one needs to become an expert in creating a number of difficult and complex data structures, for each graph and summary table type, where in JMP there are really only one or two data structures that are nearly universal for every graph and analysis report. (one observation per row, one variable per column)