Video

Great question! To do that you can use what Stata calls "factor notation." This means that you tell Stata that a given variable is categorical. To do that, use the prefix "i." before a variable. Say that your variable is called "age." You would enter "i.age" in the command and it will then create a separate line in the table for each category. This video walks gives an example starting at around the 6:00 mark.

@@jasonburns2579 Great. At that point in the clip, we get proportions in decimal places. Could I use that for mean in representing Table 1 of summary statistics say for a numerical variable coded with levels like age or BMI?

@@RosatiSamuel For that you would want to create a crosstabluation or two-way table. The idea with that is you declare your categorical variable as the row variable and then specify which descriptive statistics you want in each row. A simple way to do that would be to use the command "tabstat BMI, by(age)" You can customize it more using the table and tabulate commands. I have a video on crosstabs you may find helpful in this scenario.

@@jasonburns2579 Thanks alot.

you can use the command "tab1 [varlist]" to generate a frequency distribution table

You can download a copy of the data from my website at www.jasonburnsphd.com/new-jersey-school-performance-data but you will need your own copy of Stata

A limitation of a chi-square test is that it does not reveal whether, or which, specific value of either variable is driving the result. Looking at the descriptives could give you some insight about where differences are greater, but can't be used to draw any conclusions about what drives the p-value/difference. If you have a strong (theoretical) reason to hypothesize that specific values/groups are important then other tests could be used. For instance, one may not really be interested in counties, but just the difference between traditional public schools and charter schools. In that case, one could create some new variables that group all public schools together and all the charter schools together and then calculate the proportion of schools from each group that are high, medium, and low poverty, which would allow one to run 3 t-tests (using charter/tps as the independent variable in all three and then using proportion low poverty, proportion, medium poverty, and proportion high poverty as the dependent variables) to investigate whether charter/tps schools are more/less/equally likely to be low, medium, or high poverty.

Hi and great question! When specifying a variable as an indicator (using "i."), in the backgroun Stata then treats each value of the variable as binary where 1 means that an observation is in that group and 0 means that it is not. That is then used to calculate proportions (in the mean column) and standard deviations. Hope this helps!

@@jasonburns2579 thanks Jason, however I still don’t know how to interpret the standard deviation stats in this case.

@@mbout3456 Ah ok. Here the standard deviations can be interpreted as a standard deviation for a proportion. The text here is unfortunately a but clunky but there are some good sources with more detail on this topic that can be found online

Thanks for the feedback! I add this effect in post production to make it easier to follow the cursor, but I understand that this makes it harder to see other parts of the screen.

Without seeing the menus or errors that you are seeing, my guess is that it may have something to do with your account. It is possible that your account "tier" or "package" has limited options for customizing the appearance of your survey. FWIW, I recommend not making many changes to the appearance of your survey throughout to avoid potentially confusing or distracting respondents. Hope that helps!

@@jasonburns2579 thank you, my goal was only to make the few matrixes that I have, a bit more clear. Cause one has like 32 Items.

@@birdsandbutterflies-uu5wf Ah I see. To do that I would break the table up over a few pages. Some reasons for that are that it will help to break things up a little bit like you are trying to do. That will also make sure that you get more data. The way Qualtrics works, no responses are saved until the respondent hits the "next" button at the bottom of the page. If you have a matrix with 32 items and a person responds to 16 of them and then exits the survey, none of those 16 responses will be saved. If you have those 32 items spread across 4 pages (8 per page) and the person responds to all 8 on the first page and then goes to the next page, you will have those first 8 responses. If the person then responds to the 8 on the second page but doesn't click next, you would lose responses for items 9-16 but keep responses 1-8. If they click next on the bottom of the second page, then you'd have all 16 of those responses. Lastly, consider how your respondents will be interacting with your survey. Most of the time the way that matrix items render on mobile devices can be a little clunky in a way that I think increases response burden. Most of the time it presents each item and the scale individually rather than in grid form like it does on desktop. Just something to think about.

@@jasonburns2579 thank you SO MUCH! 1. for your kind effort to give me this wonderfully detailed and clear explanation and 2. for simply giving me this information overall, because I didnt know that qualtrics works that way. This is my first time! My problem is that those 32 items are attached to that one question. For instance "what are your favorite traits in a person" and then they can choose multiple items to answer that single question. Do you have any suggestions how to solve that problem smartly? again, thank you so much:)

@@birdsandbutterflies-uu5wf No problem! So for that kind of an item, the only way to split it up would be to maybe group traits according to some kind of theme (personality, physical appearance, etc.). Based on your description, if it is a "select all that apply" type of item then it may be ok. One comment on the stem "What are your favorite traits in a person?" I would maybe change this to ask about the traits one admires in a person or something like that. "Favorite" implies a kind of ranking that could confuse the respondent (how many favorites can one have?), but there are ranking-type items within Qualtrics. Also, think about what you want your resulting data to look like. If you use a "select all that apply" type of item, which I think Qualtrics calls multiple response, then if you export the data you will get all of their selections in one cell of the data with each thing the respondent selects separated by a comma. That can be a pain to analyze because each cell in the column for that item could in theory be different for each respondent, making it tough to do any sort of quantitative analysis. Another option would be to have a yes/no response for each characteristic, which would add some response burden (so consider how long the overall instrument is) but the resulting data will have a separate column for each characteristic and their response (yes, no, or missing), which requires a lot less cleaning prior to analysis.

Glad it was helpful!

Changing the background picture wouldn't impact data collection in a strict sense, meaning that variables wouldn't be added, deleted, or recorded differently, but it may affect some things in a relevant way. One is that in theory, a different background image may lead respondents to provide slightly different answers subconsciously (you could investigate this further by looking into modal effects in surveys). That could happen if the image distracts the respondent or if the image conveys some sort of message like a desirable response. Another is that a different appearance may lead some respondents to think that it is a different survey, though for that, context matters a lot. If, say, it is a survey of people at a workplace where many people would be invited to participate and know each other, one person seeing the new survey on another person's screen may lead them to think that they need to participate in what they think is a new survey or data collection. That part wouldn't impact the data but could wind up in a number of emails related to survey administration. Overall, a small change is probably fine and for any change it would be worth thinking through the extent to which a change in appearance may lead to any of these issues. Hope this helps!

That most likely is due to the permissions on your account. If you have an individual account, the tier of your account may not allow you to change the look and feel of an instrument, meaning you may need to upgrade in order to do that. If you are part of an organization, then you may be able to change the look and feel of a project by contacting your brand administrator and asking them to give you that permission.

@@jasonburns2579 yup yup, thank you 🙂