*However, if you cannot read my handwriting, I will be glad to translate it for you.

Although this assignment does not count very much toward your final grade, and although it is the first one of the semester, it is probably the hardest thing that you will do in Statistics this Fall. When you recategorize a variable, you need to do a lot of thinking about what that variable and its values mean. You need to decide what makes the people or other types of  cases "in the same category" the same, and what makes people in other categories different from those in the first category.

In addition, you needed to know the measurement level of a variable in order to select the correct modal and median values for it. As coded in its original form within the SDA system, "emailhr" is ratio. After your category collapses, of course, this became an ordinal variable.

If you took a mean on the recoded variable, you were in error. This recoded variable was no longer interval, ratio or numeric data. It became ordinal data after your recode.

Most assignments were quite good. Some absolutely terrific rationales that were very interesting to read. Very heavy email users were held in low regard and received labels such as "addicted" or "no life beyond the screen," Nonusers were also perceived negatively and received labels such as "the lazy" or "hermit".

Virtually all the tables were attractively drawn, typically using a word processor or spreadsheet program.

PLEASE NOTE: If your score was below 15 on this assignment, please review readings, Guides and Assignment 1 very thoroughly. Decide if you made careless errors (that you won't make again) or whether you have some serious misunderstandings of the material. A tutor may be in order. Generally, it is NOT a good idea to use a math or math statistics major because this course is NOT about formulae, but about data decisions. I can make some suggestions if you give me one session's lead time or email me. And, of course, Maria and I will be glad to help you. (So try us first.)
 
 

• Completed all parts of the assignment. CRITICAL!
• Followed all directions.
• Turned in all computer output.
• Ran the programs correctly (that was the easy part and everyone did.)
• Retained all the valid cases while recoding the data.

It is vital to be able to construct and read a well-presented table. Although the table is a basic building block of data analysis, it is used constantly in both popular and professional prose.

You constructed a UNIVARIATE PERCENTAGE table, a very widely used data presentation tool.

THE TABLE ITSELF in this assignment:

• was uncluttered because it presented only percents and not cell frequencies but it

• did include the total valid casebase. (It was OK to put the valid casebase in parentheses underneath the 100.0%.) There were a few deviations here and at least one casebase where I could not figure out the source.

• The percents themselves totalled approximately 100 percent (don’t worry about rounding errors such as 99.9 or 100.1% but do make a note of them).

• Deviations from 100.0 percent were labeled with a footnote stating deviations were due to rounding error (or you rounded the modal category up or down very slightly so that percentages added to 100.0).

• WAS COMPLETE. It had a TITLE, a DATA SOURCE, a VALID CASEBASE, and the amount of missing data. The most likely omission in this group was the missing data.

• I hope you remembered to use "n" for a sample, instead of "N" for a population (a few people did not).

• Each category was labeled in a descriptive manner and included the age values for that particular category.

YOUR CATEGORY SYSTEM:

• Clearly stated which categories it grouped together to form new categories.

• In its rationale, clearly stated WHY the particular categories were grouped together the way they were and what individuals in each category had in common.

• Assigned each new category a new descriptive label (and included the number of hours so that your reader could tell at a glance what categories such as "low" or "moderate" meant.)

• Did not deliberately create categories with either zero cases or less than one percent of the cases.

• Recognized that nonusers--persons with computers who had zero email hours were qualitatively different than individuals who used email even 1 hour a week.

• Did not try to force categories into equal intervals of 14 hours each because this created categories that were either enormous (over 90 percent of cases) or very tiny (less than 0.5% of cases).

• Did not try to force categories into a "normal curve". Data MUST BE NUMERIC TO FOLLOW A NORMAL DISTRIBUTION. As soon as you collapsed categories fromm 33 to 5, the data became ordinal, hence nonnumeric (we will resume with the normal curve on Monday).

• Did not create "intuitive" categories without sharing the reasons behind your intuition with the reader (me).

• Used ALL the valid cases.

• Used the mode (0) and median (2 hours) as directed from the original 33 values of weekly email hours.

• Applied cumulative percents correctly for the "less than 7 hours per week" question (that's 6 hours per week OR LESS--that was 81.1 percent which you could actually get directly from the "bad recode example output)  and the "at least 30 hours per week question" question (that's 30 hours per week OR MORE--that was either 2.3 or 2.6 percent depending on whether you cumulated frequencies or percentages; the differences were due to rounding).

  ---

You lost credit if:

• You did not construct a table, but substituted computer output or something else instead of a percentage table.

• You omitted parts of the assignment and did not answer them.

• You mixed up the mean value with the median value.

• You used the statistics from the recoded table instead of the original table.

• You used the cumulative percent from seven or less instead of the correct less than seven hours weekly.

• You had some cumulative percents and I could not figure out how you obtained them at all.

• You included BOTH frequencies AND percentages in your table.

• You cluttered up the table by putting a percent mark in every cell, instead of just the first category percentage and the total of 100%.

• You omitted percent marks entirely (having the % mark at the very top of the column was OK)/

• You omitted the valid casebase.

• You omitted any mention of missing data entirely, or incorrectly designated the missing cases. These were largely cases which did not have either computer or Internet access somewhere. The unselected cases for the surveys ("filtered out") for the years 1972-2000 are not missing. You simply selected only the cases for the year 2002.

• You omitted a descriptive title.

• You omitted the source of the data or you used an incorrect source. We used the year 2002 General Social Survey for this assignment.

• You miscalculated cumulative percentages.

• You put a percent, and not a category, for the mode or the median.

• You put nonusers and email users in the same category.

• You created either enormous categories (over 90 percent of the cases in one category) or very tiny categories (e.g., less than one-half percent). Very tiny categories are basically like "throwing away" a category. Sometimes you can't help this because that's how the observed data are distributed. But if YOU create the categories, these very large or small frequencies can usually be avoided.

• You used an incomplete subset of the assignment from last year's or last semester's EDF 5400 course which used a totally different variable. If by chance you route to the wrong course year or semester for EDF 5400, ALWAYS check for the course year at the top of the page and make sure it reads "Fall 2004." Also be sure that the construction sign has been removed from the top of the assignment page (or any other page for this course) before you print the website.

• You didn't understand a cumulative percentage so the answers to question 6, parts 3 or 4 were incorrect.

• Become used to slightly different ways of describing a cumulative percentage. You need a cumulative percentage to easily calculate the interquartile range (IQR), a very useful measure of dispersion for ordinal data. There WILL be cumulative percents on Exam 1.

• Your "cumulative percentage" didn't resemble any of the data.

• You didn't explicitly say which categories you collapsed to make a new category, leaving your reader to guess at what each category contained.

• You didn't tell why you grouped together the categories that you did, that is, you didn't say what it was that people in the same category had in common. And how these people differed from people in the other categories. Thus your rationale was incomplete.

Remember that individuals who did not have any computer access were coded missing. They were NOT coded zero. Individuals with no computer access were not asked any questions about email usage. If you included individuals without a computer in the "0" group, you lost credit because the assignment notes these people were coded as missing. I also mentioned this in two different classes.

Individuals with computers who were total nonusers belong in a DIFFERENT category from users, even those who use as little as 1 email hour per week. This is a qualitative difference. Further, these individuals comprised nearly one-third of the sample so the category was not negligible.

Almost all the rationales were clear and a lot of fun to read. Even when the same approach was taken (e.g., probable work use), different people used different rationales and different coding schemes.

In either case, as you go up the hourly scale, there are generally fewer people who used email a great deal. So be careful not to create categories that are so tiny that you basically "threw a category away" and really only had FOUR categories instead of five.

You will have only a limited number of categories to describe your data in a univariate distribution. Maybe you will have more than five categories--but not too many more than that because it will make the table too difficult to read. You typically don't have categories to "throw away" on tiny numbers of cases.

READINGS AND ASSIGNMENTS

OVERVIEW

Susan Carol Losh September 19, 2004
This page was built with Netscape Composer
and is best viewed with Netscape Navigator
600 X 800 display resolution.

This was Hurricane Frances