nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

case-wise or row analysis question

emeritus

I have some survey data that has descrete variables measuring item response time in sec. to completion each questions. I would like to isolate respondents who are consistently taking a lot of time across all questions. Say a respondent takes more or less 30 secs or more to answer each question. Any ideas?

Accepted answers

gzmorgan0

An alternate approach is:

Stack the item response columns, hopefully each column name identifies the Item/question
The resulting table should have a column of respondent, Label (item col name) and Data (time)
Here you might want to create a new column that is the z-score (standardized) time for each question, or at least the deviations from each questions median time [ to do this in 2 steps, create a column of the mean or median time By Label (item), then create a 2nd column that is :Data - {new column).
Now you can get a display of Standardized or Deviations by respondent, this will display both mean and standard deviation. Note using Graph Builder you could sort the nominal X-axis by the Median.

There are multiple metrics, typically, order statistics (quantiles), to select/order/find the respondents with large decision time.

All comments

txnelson

Assuming the response time columns are in the same data table as the response columns, then you could easily create a new column which is the average of the all of the response time columns. Then using a Data Filter, you could eliminate all of the respondents that have a mean time over 30 seconds.

emeritus

I think I see what you are getting at. When you say create a new column
that is the average of all RT columns, would that yield an average RT for
the respondent cross that respondent's row? That might get a little
tricky. I have a number of feeling thermometers asking about liking for
Trump and Obama, etc. typically a respondent will rank one very high and
the other very low, with an average of 50 on a 100 point scale. What I am
trying to do is cull out respondents that may be trolling the survey one
way or another. There is a variable showing the total time it took to
complete which is useful but not quite what I want. I am thinking I want
to see a trend analysis, not knowing quite what that is as its outside my
range of expertise.
BTW I have discovered the local filter and like it a lot.
Thanks
John

gzmorgan0

An alternate approach is:

Stack the item response columns, hopefully each column name identifies the Item/question
The resulting table should have a column of respondent, Label (item col name) and Data (time)
Here you might want to create a new column that is the z-score (standardized) time for each question, or at least the deviations from each questions median time [ to do this in 2 steps, create a column of the mean or median time By Label (item), then create a 2nd column that is :Data - {new column).
Now you can get a display of Standardized or Deviations by respondent, this will display both mean and standard deviation. Note using Graph Builder you could sort the nominal X-axis by the Median.

There are multiple metrics, typically, order statistics (quantiles), to select/order/find the respondents with large decision time.

emeritus

Thanks, this was a big help