ASA Connect

Hi, Andrew:

Not this exact problem, but indirectly. There are many free online stats resources, open textbooks, and so on.  What makes it difficult to offer them is that I have to be sure they use exactly the same definitions, recommendations, and formulas that I use. And that they cover things at roughly the same depth, Otherwise the students get confused.

We statisticians do not always agree on basic things.  For example, I don't teach two-sample z tests for means at all. Since t-tests do the job for any sample size, I see no point to burdening students with another test.  Other instructors may make the z test primary and present t as a small-sample alternative.

There are different approaches to Tukey's outlier test.

Excel will calculate a 25th percentile differently than you get by hand. SPSS will do an interpolated median, but it isn't the one seen in text books. z tables come in different formats. Some people insist on the technically-correct definition of a confidence interval, whereas others (like me) teach that but use a more flexible interpretation.

One time a speaker at the ASA insisted that we define the range as the difference and not the end points. I tell my students technically it's the difference but now mostly forget I said that and provide the two values!

I have my students use my notes. I provide a LOT of exercises and sample exams. I tell them the exact form of the answer on my online quizzes ("Your answer will be a decimal, not a percent, like 0.234"). 

I agree that working with online materials is tricky!

Ed

Hi, Andrew:

Not this exact problem, but indirectly. There are many free online stats resources, open textbooks, and so on.  What makes it difficult to offer them is that I have to be sure they use exactly the same definitions, recommendations, and formulas that I use. And that they cover things at roughly the same depth, Otherwise the students get confused.

We statisticians do not always agree on basic things.  For example, I don't teach two-sample z tests for means at all. Since t-tests do the job for any sample size, I see no point to burdening students with another test.  Other instructors may make the z test primary and present t as a small-sample alternative.

There are different approaches to Tukey's outlier test.

Excel will calculate a 25th percentile differently than you get by hand. SPSS will do an interpolated median, but it isn't the one seen in text books. z tables come in different formats. Some people insist on the technically-correct definition of a confidence interval, whereas others (like me) teach that but use a more flexible interpretation.

One time a speaker at the ASA insisted that we define the range as the difference and not the end points. I tell my students technically it's the difference but now mostly forget I said that and provide the two values!

I have my students use my notes. I provide a LOT of exercises and sample exams. I tell them the exact form of the answer on my online quizzes ("Your answer will be a decimal, not a percent, like 0.234"). 

I agree that working with online materials is tricky!

Ed

As Edward Gracely points out, students expect a formula to yield "the" answer (e.g., quartiles). A slight difference arising from alternative definitions can be frustrating. Even Excel has two formulas for quartiles (QUARTILE.EXC and QUARTILE.INC) and percentiles (PERCENTILE.EXC and PERCENTILE.INC). For a deeper dive into the seemingly simple definition of a "quartile" see Freund, John E., and Benjamin M. Perles. "A New Look at Quartiles of Ungrouped Data." The American Statistician 41, no.
3 (August 1987), pp. 200–203.

Hi, Andrew:

Not this exact problem, but indirectly. There are many free online stats resources, open textbooks, and so on.  What makes it difficult to offer them is that I have to be sure they use exactly the same definitions, recommendations, and formulas that I use. And that they cover things at roughly the same depth, Otherwise the students get confused.

We statisticians do not always agree on basic things.  For example, I don't teach two-sample z tests for means at all. Since t-tests do the job for any sample size, I see no point to burdening students with another test.  Other instructors may make the z test primary and present t as a small-sample alternative.

There are different approaches to Tukey's outlier test.

Excel will calculate a 25th percentile differently than you get by hand. SPSS will do an interpolated median, but it isn't the one seen in text books. z tables come in different formats. Some people insist on the technically-correct definition of a confidence interval, whereas others (like me) teach that but use a more flexible interpretation.

One time a speaker at the ASA insisted that we define the range as the difference and not the end points. I tell my students technically it's the difference but now mostly forget I said that and provide the two values!

I have my students use my notes. I provide a LOT of exercises and sample exams. I tell them the exact form of the answer on my online quizzes ("Your answer will be a decimal, not a percent, like 0.234"). 

I agree that working with online materials is tricky!

Ed

I'm  not too worried about how quantiles are calculated. We discuss the different ways things can be done. 

I'm more worried about consistency of the system. If they will use a 2 sample z-test, use it consistently.

A student I worked with had an issue on his HW. He used Excel, as instructed by the prof, to calculate an amount of money owed after X years of payments. He got something like $186,243.22. The system said he was wrong. The answer it wanted was $186,243.21. Excel uses double float calculations. The answer used float point calcs. How is a student who thinks Excel is complicated programming going to know to change from double float to float point calcs? 

How would any of us feel if our child failed a math class or did really poorly in a math class because the online system was just plain horrible? 

How would we feel if our child was marked wrong because they properly rounded a number. Whereas, the computer system truncated the number? 

I can understand a system wanting a fraction (4 1/3) over say a decimal number (4.33333). But 13/3? This is just bad.

As Edward Gracely points out, students expect a formula to yield "the" answer (e.g., quartiles). A slight difference arising from alternative definitions can be frustrating. Even Excel has two formulas for quartiles (QUARTILE.EXC and QUARTILE.INC) and percentiles (PERCENTILE.EXC and PERCENTILE.INC). For a deeper dive into the seemingly simple definition of a "quartile" see Freund, John E., and Benjamin M. Perles. "A New Look at Quartiles of Ungrouped Data." The American Statistician 41, no.
3 (August 1987), pp. 200–203.

Hi, Andrew:

Not this exact problem, but indirectly. There are many free online stats resources, open textbooks, and so on.  What makes it difficult to offer them is that I have to be sure they use exactly the same definitions, recommendations, and formulas that I use. And that they cover things at roughly the same depth, Otherwise the students get confused.

We statisticians do not always agree on basic things.  For example, I don't teach two-sample z tests for means at all. Since t-tests do the job for any sample size, I see no point to burdening students with another test.  Other instructors may make the z test primary and present t as a small-sample alternative.

There are different approaches to Tukey's outlier test.

Excel will calculate a 25th percentile differently than you get by hand. SPSS will do an interpolated median, but it isn't the one seen in text books. z tables come in different formats. Some people insist on the technically-correct definition of a confidence interval, whereas others (like me) teach that but use a more flexible interpretation.

One time a speaker at the ASA insisted that we define the range as the difference and not the end points. I tell my students technically it's the difference but now mostly forget I said that and provide the two values!

I have my students use my notes. I provide a LOT of exercises and sample exams. I tell them the exact form of the answer on my online quizzes ("Your answer will be a decimal, not a percent, like 0.234"). 

I agree that working with online materials is tricky!

Ed

Hi, Andrew:

Not this exact problem, but indirectly. There are many free online stats resources, open textbooks, and so on.  What makes it difficult to offer them is that I have to be sure they use exactly the same definitions, recommendations, and formulas that I use. And that they cover things at roughly the same depth, Otherwise the students get confused.

We statisticians do not always agree on basic things.  For example, I don't teach two-sample z tests for means at all. Since t-tests do the job for any sample size, I see no point to burdening students with another test.  Other instructors may make the z test primary and present t as a small-sample alternative.

There are different approaches to Tukey's outlier test.

Excel will calculate a 25th percentile differently than you get by hand. SPSS will do an interpolated median, but it isn't the one seen in text books. z tables come in different formats. Some people insist on the technically-correct definition of a confidence interval, whereas others (like me) teach that but use a more flexible interpretation.

One time a speaker at the ASA insisted that we define the range as the difference and not the end points. I tell my students technically it's the difference but now mostly forget I said that and provide the two values!

I have my students use my notes. I provide a LOT of exercises and sample exams. I tell them the exact form of the answer on my online quizzes ("Your answer will be a decimal, not a percent, like 0.234"). 

I agree that working with online materials is tricky!

Ed