Doing research by definition means that you are moving to new territory, that is why it is very hard to present ready made systems that can be applied to many situations.
If you are working with FACS or MARTEMEO Codes or any other fixed system, this can be done, using the predefined definitions. But most projects require very specific setups, due to the nature of doing research.
In an attempt to lend a generic helping hand, please pay attention to the following rules and tips to get going:
oGenerally, a behavioral Code should only contain a single kind of information.
oLimit your list of Codes to choose from per pass - when more than 15 Codes or so are listed in the same view the risk of making errors increases. And although the number of Codes, to be used in INTERACT, is unlimited, it is nearly impossible to concentrate on so many behaviors in one pass.
oStart with a set of 'simple' Codes like the physical situation or the phases of the observed scene, to get to know the material, before you start coding behavior that is more difficult to observe.
oCode behavior you cannot accurately observe simultaneously in separate coding sessions (multiple passes through the video).
oTry to use mutually exclusive Codes whenever possible; this simplifies the coding process.
oUse the possibility to define multiple ‚strings’ of mutual exclusive Codes per Code Definition file.
oWe recommend to store only Codes, that can be observed in one pass into, a singleWe recommend storing only Codes you can observe simultaneously in one pass into a single Code Definition file (less to choose from = fewer mistakes).
oPer coding-session, you can start with a different Code Definition file.
oEvents from different codings sessions can be collected inside the same data file AND within the same DataSet, even if they belong to different Code-Definitions.
oFor multi-level coding systems: Only first level Codes can be used to create Events.
oSub-level Codes can only be used, to further specify already logged behavior.
oIndependent first-level Code Definitions, can share the same Sub-level Code definitions.
oAll Codes that can be applied to the same Event require a different Class.
oEach partly overlapping behavior requires a separate Event.
oMutual exclusiveness works best in coding mode Standard (ad-hoc).
oYou can mix durational with none durational behavioral Codes on the first level only.
oSub-level Code Definitions are opened automatically if the coding mode Lexical is active.
oComplex behavior, that requires multiple Codes to describe it, require a multi-level Coding system. In this case, each Event holds multiple Codes in different Classes (e.g. "verbal expression" in Class "communication" + "Therapist" in Class "Recipient" + "vague" in Class "Quality").
oA multi-Level Coding System requires a hierarchical set of Code definitions that are linked by lexical chains. That way you do not need to switch between the levels manually. This type of code definitions require the Lexical (post-Hoc) coding mode.
Note: It is always easier to combine single Codes afterward than extracting information from a combined expression!
If, at the end, you do need to aggregate some of the differentiated Codes, to extract the required information, INTERACT offers several Transformation routines. Very helpful in most cases, is the routine Move & combine Codes.
Arrangement of Codes, Classes and DataSets
There are always different views on the same information to be collected. This has immediate impact on the structure of the coding system and vise versa. Below the basic variations:
 | Sessions - Typically, each session is stored in a separate DataSet. Read Structure logged Events for details. |
| Data per Type - Create a separate Class (=column) for each type of behavior (E.g. Class 'Communication' with Codes 'Verbal', 'Gesture', 'Mimic', 'None', etc.). This enables you to use mutually exclusive codes. Different subjects can be coded in separate DataSets, different sessions in separate DataGroups. This structure offers statistics per behavior, per subject but NO co-occurrences or other relations between behavior of different subjects. |
| Data per Subject' - For a very limited number of Codes, creating a single Class (=columns) per subject/object, can work as well. Create a Class per subject (Classes: "subject 1", "subject 2", "subject 3") In this case, all Codes (e.g. 'back', 'front', 'left', 'right') are required for each Class. Ideal only if you can again make use of mutually exclusive Codes. This structure offers statistics per behavior and subject AND co-occurrences / relations between all behaviors is possible. |
| Data per Type and Subject' - Requires Multi-level Codes and results in multiple Codes per Event. Using the Lexical coding mode, you can create a Class called e.g. "Subject" in which you collect the Codes: 'Subject 1', 'Subject 2' and 'Subject 3'. Additional Classes hold different types of behavior like 'Communication' (with Codes 'Verbal', 'Gesture', 'Mimic', etc.), 'Gaze', 'Movement', etc. resulting in different Codes for each Class. This structure provides statistics per Code over all subjects at a glance. To identify the totals of each behavior per subject, you can split your data into a separate Set per Subject, using the Restructuring command Group all Events.... To analyze specific types of interaction between subjects, the State-Space-Grid analysis shows you all co-occurrences for the selected classes at a glance, for Contingency and Sequence Analysis on behavior of different subjects you need keep the data for all participants in the same DataSet and you might need to run the Move & Combine Codes routine to create the combinations you are interested in. |
| Combined Classes - Another multi-level coding system. This approach creates context based classes while coding. It collects the Codes per Class (or type), but with a separate column for each 'subject_type' combination (e.g. '1_Gaze', '2_Gaze', '3_Gaze', '1_Movement', '2_Movement', etc...). This can result in a huge number of Classes, but fortunately you do not need to define them all by hand. This structure offers statistics per behavior and subject at a glance, but an overall result per behavior, independent of the subject is not available. The State-Space-Grid analysis can ONLY be used on part of the data. Read Prefix - Create Context based Classes for further details. |
Which arrangement of Codes, Classes, DataSets and/or logging method is best for you, can, unfortunately, not be generally advised and needs to be investigated for each project separately. If you do have problems or questions, just write us an email with as much detail you can think of: support@mangold-international.com
* A 'subject' can be a test person, a focus animal or anything else that is being observed, like e.g. an area for which changes are counted.
•