INTERACT offers various ways to log your Codes. The way you want to collect data, the number of Codes needed, and the research questions you want to answer, largely influence what observation method, coding mode and coding routine is best for you.
You can use a deductive or inductive approach or even a combination of the two in different stages of your observations.
In INTERACT, you can use and combine all standard observation methods:
oEvent sampling
oContinuous sampling
oTime sampling
oPinpoint sampling
oOne-zero sampling
oAd libitum sampling
oInstantaneous sampling
oScan sampling
o...and variants
Note: The publication of Sampling Methods in Behavior Research from Philip N. Lehner in 1991 provides a nice overview of the pros and cons of the various methods.
In INTERACT we narrow down this great variety of methods to their core, so we differentiate between the following methods:
 | Event-based (Event Sampling, both with and without duration) - With this preferred observation method, Events are created whenever they occur. Event-based data is more precise and offers accurate statistics on frequencies and duration. Multiple overlapping events can be created simultaneously as often as needed, but in real life, it is usually hard to concentrate on more than 2-3 different types of behavior at the same time. Multiple passes through the same video are possible and often required. With INTERACT it is also possible to export Event-based data in Intervals, to make the data comparable with older time-sampling-based studies. |
| Interval-based (Time Sampling) - Fixed intervals are created before you start an observation session. Those empty lines are also called 'Events' and can be used to watch the corresponding part of the video and be filled with Codes, describing the situation for each of those intervals. You can use this method for zero-one sampling or other interval-based observation methods. Data collected this way is less accurate than Event-based data, because neither the frequency nor the duration of the collected Codes reflect reality, so we advice to use this method only to collect information that cannot be linked to an actual event. All Codes that can be applied to a single interval simultaneously, do require a separate Class. How to create fixed intervals is described in Create Fixed Intervals. |
Note: It is possible to export Event-based data into any required interval-width!
| Pinpoint (fixed event distance) - This method is only useful if you are logging codes without duration. It is a Time Sampling method during which you create the Events on the fly, like is done for Event Sampling. But, instead of playing the video or replaying a predefined interval, you manually log Time-Stamp-Events, every so many seconds or minutes, with the help of the Step-Width functionality. Everything else works the same as for Event-based coding. How to define your step-width, is explained in the INTERACT Help file in Step width, Pre-roll and Search Speed. |
Note: You can combine all methods within the same data file, but you can use only one method per observation session.
To combine methods, you need separate passes through your video.
Example: Code behavioral Events as they occur in one or more passes to get the most accurate data collection. Additionally, you can create predefined interval-Events to judge the emotional state; both types of Events can be stored within the same DataSet, independent of the processing order.
But per observation pass, you can either create new Events OR log Codes into existing (interval-)Events.
The chosen method does influence the logging process and might require an adaption of your Code definitions.
Time Sampling vs. Event Sampling
Studies from the 70ties and 80ties are based on the interval-based Time-Sampling coding method, because there was hardly any other way to do this (using a VCR and pencil and paper), but today you can get much better results with about the same amount of labor, using an Event-based coding method.
The trouble with time-sampling or interval-based coding is that neither the duration nor the frequencies are correct.
Example: Let's say you work at 5 second intervals.
If a behavior begins at the last two seconds of "Interval 1", then lasts for the entire "Interval 2" and finally ends in "Interval 3" after the first 2 seconds, this statistically gives you a frequency of 3x and - in this example with an interval of 5 seconds - a duration of 15 seconds.
While in real life the behavior occurred only once and lasted only 9 seconds instead of 15 seconds!
Performing Event-bases observations with INTERACT, using mutually exclusive codes whenever possible, might even be easier than performing an interval-based observation and it definitely results in better data.
If you need to compare your results with older studies, you can export the event-based data from INTERACT to any interval grid to get the best of both worlds!
During each interval-based export, you can even decide what percentage of the Interval needs to be covered by a code to be labeled ‘present’.
| TIP: | If you are not interested in the duration for some of the Codes, you can also mix duration and none-duration Codes within the same Coding System. |
Maybe even more important is the fact that for Interval-based data collection, Codes that could co-occur within the same interval need to be stored in separate Classes. This often results in Classes with only a single Code.
This 'single Code per Class situation' is a problem for calculating inter-rater-reliability, simply because the kappa calculation is primarily based on probability mathematics, and that doesn't work well for a single variable.
There are only a few situations that are better off with interval-based observations, so I hope you reconsider your approach. It is just a pity to have a great tool at hand and only use it in an old fashioned way. We will be more than happy to help you streamline your coding system for Event-based coding if you can spare the details.
Coding Routine
How you best log your Codes, largely depends on the number of Codes and the nature of the behavior you are looking for.
oComplex, multi-level Codes require another approach than simple yes/no situations.
oOften, it is best to work with different passes through your material, using small, separate Code definitions per pass.
oSometimes even a complex looking coding scheme can still be easily coded in one pass.
Sequential vs. Incoherent
Logging sequential Events (each Event is seamlessly followed by another Event), is often easier and faster than logging independent, incoherent Codes. Even if it means that you need to add an extra Code, to log all situations you are NOT interested in. Also, sequential Events give more reliable results for most analysis routines.
| Mutually Exclusive Codes - The fastest way to log sequential Events, is to use mutually exclusive Codes. Logging Events with mutually exclusive Codes means that you only need to log the start of each new behavior. INTERACT allows you to log multiple 'strings' of sequential Events simultaneously, if you use different exclusion lists. The next behavior (from the same exclusion list) automatically closes the previous Event. For this, the Standard coding mode, in combination with at least one exclusion list, is perfect. |
| Turn by Turn Events - The easiest way to log sequential Events, is logging the start and end of each Event with the SPACEBAR. The Coding Mode Lexical allows you to do just that, saving you the trouble of looking into the future. At the end of each Event, the video pauses and you have time to select one or more codes to describe the just logged Event. You can enter multiple Codes per Event. |
| Parallel Strings - The most complex way to log sequential Events, is to log sequential Events for multiple types of behavior or subjects simultaneously. This can be done using the Complex coding mode as well as with the Lexical coding mode. Often, it is advisable to use one of the simpler routines, during multiple passes through the video. |
| Incoherent Events - If you want to log occasional Events, that appear randomly, leaving gaps between Events and creating random overlaps, you always to need to log the start and the end time for each Event manually. This is done by clicking the same Code at the start and at the end of the behavior. For rather short events, the Push & Release coding mode is helpful, because you do not need to click twice; you just hold the Code pressed for as long as the behavior lasts. |
Remember: Any of the examples can be adapted to a completely different field, because the difference lays not so much in what you observe, but in the way you observe!