When users ask us how to code intervals in INTERACT, our first return question is: Why do you want to do interval-based coding?
There are a few situations where coding intervals (Time-Sampling) is indeed the best solution. But often it is not, especially because Event-based data is far more accurate.
The two most common reasons we get as an answer:
1.Because my reference study was performed that way.
2.To get better Reliability results.
With INTERACT you can have the best of both worlds:
Collect high-quality Event-based data for new insights with the ability to export your Event-based data in any interval length you need for comparison with previous results.
We hope you at least re-think your reasons and consider a different approach for better in-sights due to more accuracy. In some cases, there is no way around time-sampling though, which is fine, if done for the right reasons.
Reference Studies
Many studies find their origin in the 1970's/1980's. Back than, all observations were either live or based on VCR recordings. At that time, the only practical way of collecting observations, was a predefined set of intervals to observe and add information to.
But collecting your observations in Intervals results in less accurate data, because neither the Duration nor the Frequency of the collected information is correct!
Example: Presume we use 5 second intervals.
If, e.g., a behavior starts at the 2nd second of the first interval, lasts over the full five seconds of the second interval and ends at the 3rd second of the third interval, its Code is entered for all three intervals.
This results in a frequency of 3 for that Code and a duration of 15 seconds. but in real live, the Code occurred only once and the actual behavior is 11 seconds, not 15.
The difference between 'observed' and 'occurred' behavior increases if the behavior reoccurs the more often or the interval is even bigger.
Using a modern tool like INTERACT allows you to collect high quality Event-based data with the least possible effort.
Using mutually exclusive Codes or a clever hierarchical setup can simplify your coding live a lot. Ask us for assistance at support@mangold-international.com.
After data collection, you can export the data in any interval required for comparison with any interval-based studies.
Reliability Results
Kappa - For all text-based Codes, only the Kappa Inter-Rater-Reliability calculation is suited. The original formula was developed for mutually exclusive and exhaustive Codes only.
Because typical observations often overlap and might include gaps, we can only calculate Kappa per Class and we developed a unique Pair finding routine to identify matches, based on which the original Cohen's Kappa formula can be applied.
Parameters allow you to specify percentage of overlaps as well as allowed offsets for very short behaviors. These Kappa Parameters usually eliminate most of the accuracy issues/questions.
Although it is not possible to turn 'bad data' into 'good data', it certainly is possible to make the best of the data collected. There is no general 'perfect-match' setting, it totally depends on the required accuracy, the length of your codes, and the difficulty to detect certain behavior. You might even need to run the routine with different settings for different Classes.
The best thing of our Kappa is, that the Kappa-Graph visualizes the effect of your settings, enabling you to understand the flaws. A cross-reference between the Kappa-Graph an the original data file allows you to jump to a problematic Event with a double-click.
Draw-Backs of Time-Sampling
Coding fixed intervals may seem easier, and sometimes that might even be true, but it comes with a loss of many details:
oThe resulting Frequencies of the descriptive statistics do not match reality.
oThe calculated Duration in the descriptive statistics is not accurate.
oSequence Analysis cannot find sequences among Codes, only among intervals, resulting in many repetitions for each repeating Code.
oContingency Analysis can only look for the next segment, because for repeating Codes the start time is automatically set to that of the current interval.
Turn Intervals into Events
If you do have time-sampling data and are now struggling with one or more issues listed, INTERACT can turn your intervals into Events!
The command Transform - Events - Merge 
allows you to do just that.
▪For details on this command, read the topic Merge Successive Events.