<< Click to Display Table of Contents >> Navigation:  ANALYSIS with INTERACT > Lag Sequential Analysis

Note: The INTERACT Sequential Analysis module requires an additional license.
If you cannot find the command Sequential analysis in the menu Analysis - Evaluation - Statistics or the Analysis menu of the Timeline Chart window, the function is not activated in your INTERACT version.
Contact our sales team at sales@mangold-international.com, to learn how to you can obtain this additional feature.

Lag Sequential Analysis (LSA), is a technique for analyzing quantitative data, in which the sequential order of the data is examined. Originally published as the Allison and Liker method from 1982. In INTERACT we use the adapted version of Roger Bakeman and Thomas M. Gottman.

It is traditionally used in the field of developmental psychology, to study the behavior of person to person interaction (how often certain behavior is preceded or followed by other specific behavior) with the intention to accurately describe the collected data.

LSA can also be used for other tests, for example during usability studies, to identify typical errors or to trace operating progression.

With Lag Sequential Analysis, evaluators can generate statistics, that capture aspects such as frequency and conditional probabilities of behavioral Events.

This LSA routine was developed for mutually exclusive and exhaustive Codes. Sequences are counted from start-time to start-time.

In INTERACT, the Lag sequential analysis command is available directly in the Analysis - Statistics menu or from the Analysis menu within the Timeline Chart window.

IMPORTANT: Select only relevant Codes and Classes!
 
Running the LSA-routine from within the Timeline Chart allows you to limit the data by time:
only the Classes and DataSets visible in the Timeline Chart are analyzed!

You can select only specific Codes and search for sequences among those Code alone.

oHow to exclude certain Codes, Classes and/or DataSets is explained in Data Selection - Codes & Classes.

The sequential analysis method used, was designed for data in which consecutive Codes may repeat (equal repetitive Codes are possible).

IMPORTANT: If your data relies on structural zeros on the diagonal, because consecutive Codes may NOT repeat, the expected frequencies cannot be computed with this method! For this type of data, it is best to use a program called GSEQ, which offers an import routine for INTERACT data files.

Lexical Codes

If you coded multiple Codes per Event, those Codes automatically overlap - even per Event. For this type of data (in its original state), you can only identify sequences per Class. This means you need to visualize only one Class in the Timeline Chart. To combine the Codes from all Classes within a new Class column, so you can analyze all your combined data, use the routine described in Move & combine Codes:

•Click Transform - Codes - Move and Combine to start this routine, select all relevant Codes and create a new Target Class.

•Select Analysis - Evaluation - Statistics >Lag Sequencial Analysis from the toolbar and select only your new <Target Class> for analysis.

Alternatively, you can first visualize your data to be analyzed in the Timeline Chart:

•Click Analysis - Evaluation - Timeline Chart ; select only your new <Target Class> for analysis.

•Run the Sequence Analysis routine from the Analysis menu within the Timeline Chart window.

Multiple DataSets

Sequences are computed per DataSet only, independent of the fact if they overlap in time.

Run Sequential Analysis Routine

Note: Make sure analyze ONLY the Codes that are logically of interest for the sequence analysis process.

▪Open your data file.
To open a demo filer select Start -Files - Open > Demos   > Documents > Document: Sequence Analysis demo.

▪Select Analysis - Evaluation - Statistics >Lag Sequencial Analysis from the toolbar.

In this sample data, all data is mutually exclusive and can be analyzed as a whole instantly.

If your data contains multiple Classes and overlapping Codes, probably NOT all data must be analyzed at once.

▪Select only the applicable Codes.

▪Confirm your selection with OK.

The Sequence Analysis settings dialog appears:

▪Verify your search details as explained in Accuracy of Calculation.

▪Confirm your choices with OK.

Note: The routine searches for start times only!

Explanation: During the sequential analysis, INTERACT examines all available codes chronologically from start time to start time, ignoring all offsets.
The routine identifies the first start time in the selected data pool, notes the corresponding Code, and searches for the subsequent start time.
That subsequent Code is counted as a sequence of the previous Code. After that, this subsequent Code is memorized, and the search for the next start time continues, and so on.
 
If there are overlapping Codes with the exact same start time, it is NOT a sequence and only one of the Codes is counted, that is why it is important to know what you are looking for and to analyze only Codes that make logically sense for the analysis.
- Usually, you'll need to analyze your data Class by Class.
- Overlapping Events that should be analyzed as their combined behavior need to be turned into sequential Events with help of the Create Coding Segments routine.
- Lexical Codes need to be combined to an aggregated Code with the Move & Combine routine.

Sequences Found

INTERACT creates the following matrix on the tab Sequences:

The 'Results' matrix contains the plain counts on how often a certain Code is followed by another, starting from the start-time of the preceding Event until the next start-time found. Those values are used to calculate the resulting probabilities and z-values.

oAll available Codes are listed twice: once as row labels and once as column headers.

oHorizontally, the values represent the number of times the Code in the row was followed by the Code in the header, according the chosen lag. Per default lag = 1.

oCounted is always the first* Code found, since the previous Code has started, within the same DataSet - independent the Class and independent of the fact whether the previous Code has ended yet!

oThe resulting values are accumulated over all selected DataSets.

oVertically, the Codes in the column header are preceded by the Code in the row, as often as listed.

oIf you compare the total number of Events (N) to the total summary of the transitions, you see, that the total is always N - lag. In this case 137 Events - lag 1 = 136.

IMPORTANT: To get reliable results, do only apply the Sequential Analysis routine to mutual exclusive data. If your coding system is not mutual exclusive, but some of the Codes are, you can apply the Sequential Analysis routine on those Codes only. Events that do overlap in the Timeline Chart, may falsify the results.
Make use of the Pattern segmentation routine, to create sequential segments from overlapping behavior.

*) you can also search for the second (lag 2) or third code (lag 3) by adapting the Sequence analysis settings within the Timeline Chart settings dialog.

Visual Data Exploration

You can verify the results instantly, because

INTERACT offers a special visual data exploration routine, which is active per default:

▪The icon Visual Data Exploration in the toolbar, allows you (de)activate this functionality.

All results are analyzed and colored in 25% bins, like this:

For this, all values are compared per column, and the range, between the lowest and highest value, is used to split the results into four groups:

•Color 1 is used for the values within the lowest 25%

•Color 2 is used for the values within the next 25%

•Color 3 is used for the values within the third 25%

•Color 4 is used for the values within the highest 25%

This gives you immediate visual feedback on your results.

Specify Colors

You can specify the colors for quartered result sets in the application settings:

▪Select General - Program - Settings and switch to the tab Behavior - Colors.

▪Change the colors For  quartered result sets to your liking: