4. Transform data into revelatory statements.
Once you have partitioned your natural language data according to meaningful units of information, you are ready to transform it into statements that reveal its main themes. In Giorgi’s phenomenological approach, this is accomplished by the free imagination1 and explication of researchers (and perhaps key informants) involved in the study. In Raven’s Eye, this data transformation is accomplished algorithmically, by calculating the degree of association between words as they appear in their original context. By doing so, we accomplish the same goal as Giorgi (revealing insightful and contextualized examples of the meanings being relayed in the natural language sample), while dramatically increasing the reliability and replicability of results acquired.
4.1 Construct concepts by elaborating overrepresented gists.
As with word inflections discussed in 2.1, if a specific word is highly overrepresented in your natural language data, several of its synonyms and associated terms will generally also be overrepresented. Similarly to inflections, if your sample is sufficiently large and diverse, you should find clusters of such words present in your natural language sample. These clusters of words and each of their respective lexemes and determiners serve to elaborate on your main gists. When considered as a whole, we call a cluster of elaborated gists a concept.
Construct concepts by inspecting your tables for synonyms and other terms associated with your overrepresented gists. Scrolling downward in the table (or resorting it alphabetically) will generally reveal the presence of these. To verify that a synonym or other word is utilized in the same manner as the gist in question, select the synonym or other word in the table, and then inspect the use of the word via the CORVID, Word Association, or Word-In-Context features of the Advanced dropdown menu. Concepts are labeled according to the most proportional of those overrepresented words that constitute their cluster (and, from 2.1, it’s most proportional overrepresented inflection) in the natural language sample. Continue to construct concepts until you have accounted for the gists overrepresented in your natural language pool.
Construct concepts by inspecting your tables for synonyms and other terms associated with your overrepresented gists. Scrolling downward in the table (or resorting it alphabetically) will generally reveal the presence of these. To verify that a synonym or other word is utilized in the same manner as the gist in question, select the synonym or other word in the table, and then inspect the use of the word via the CORVID, Word Association, or Word-In-Context features of the Advanced dropdown menu. Concepts are labeled according to the most proportional of those overrepresented words that constitute their cluster (and, from 2.1, it’s most proportional overrepresented inflection) in the natural language sample. Continue to construct concepts until you have accounted for the gists overrepresented in your natural language pool.
4.2 Associate overrepresented concepts.
To associate an overrepresented concept, select it by placing a check-mark next to it’s main gist in either the main table, or the subsample comparisons table. Next, select the Advanced button in the main menu. Briefly inspect the Word-in-Context window to get a sense of the way in which the selected concept is contextualized in the original natural language data. Then, select the Word Association feature from the drop-down menu. Select to analyze the 5-10 words surrounding the target word, and inspect the table that appears. This will provide a general sense of the relations between concepts in the natural language sample, while step-wise reduction in the number of words included in each analysis will serially reveal those words that are more intimately related to the selected concept. Repeating this procedure for each of the gists that comprise a given concept will increasingly refine one’s understanding of the way in which the various concepts are related.
To determine the ordered relations of concepts, first narrow the Word Association to focus solely the words preceding the target word. Then, switch the Word Association feature to the words following the target word. This will reveal the ordered relations and contingencies between gists. As needed, continue refining these associations through a step-wise reduction in the number of words included in each analysis. Sometimes, when the natural language sample is highly structured (as is often the case in written responses to a specific survey question), one can easily identify associations between concepts by simply serially selecting overrepresented gists, then recording the most frequent gist that appears in the table when selecting to analyze the one to two words that following the target word in the Word Association feature.
To determine the ordered relations of concepts, first narrow the Word Association to focus solely the words preceding the target word. Then, switch the Word Association feature to the words following the target word. This will reveal the ordered relations and contingencies between gists. As needed, continue refining these associations through a step-wise reduction in the number of words included in each analysis. Sometimes, when the natural language sample is highly structured (as is often the case in written responses to a specific survey question), one can easily identify associations between concepts by simply serially selecting overrepresented gists, then recording the most frequent gist that appears in the table when selecting to analyze the one to two words that following the target word in the Word Association feature.
4.3 Construct revelatory statements based on the associations between overrepresented concepts.
Based on the context in which your concepts are used, and the associations between them, construct revelatory statements for each part (or meaning unit) of your data. If, through your previous use of the Word-in-Context and Word Association features of the Advanced tab, the association between concepts are not already apparent, the CORVID feature will generally help to reveal or confirm them. As with the other features of the Advanced tab, the CORVID feature works on the word that is currently selected in the main table. Multiple concepts can be serially analyzed by successively deselecting the current word, and then selecting a new word prior to clicking the Update button in the Advanced tab.
When calculated in Raven's Eye, CORVID creates a table that depicts a network (or tree) of ordered relations between words in your natural language pool. It does so by returning two branches: one consisting of the most frequent words found preceding the selected word as it is used in your natural language sample, and one consisting of the most frequent words that were found to follow the selected word as it is used in the natural language sample. Each branch, in turn, consists of two orders. The first-order is comprised of the top three most frequent words that directly precede or follow the selected word as it is used in your natural language sample. In the table, the columns containing first-order word associations are labeled 1st word. The second-order is, in turn, comprised of the top three most frequent words found to be directly preceding or following each of the three words listed in the first-order. In the table, the columns containing second-order word associations are labeled 2nd word.
Through this network, CORVID automatically delineates the relationships between concepts as they are associated in the natural language sample being analyzed. Connecting together words across orders in the CORVID feature creates aptly labeled revelatory statements, which both reveal and represent popular means of relating concepts in the natural language sample (and, by extension, the words and thoughts of the person or people supplying it). These revelatory statements, therefore, relate the main concepts of your natural language sample in words and ways that maintain high degrees of fidelity to their original expression.
When calculated in Raven's Eye, CORVID creates a table that depicts a network (or tree) of ordered relations between words in your natural language pool. It does so by returning two branches: one consisting of the most frequent words found preceding the selected word as it is used in your natural language sample, and one consisting of the most frequent words that were found to follow the selected word as it is used in the natural language sample. Each branch, in turn, consists of two orders. The first-order is comprised of the top three most frequent words that directly precede or follow the selected word as it is used in your natural language sample. In the table, the columns containing first-order word associations are labeled 1st word. The second-order is, in turn, comprised of the top three most frequent words found to be directly preceding or following each of the three words listed in the first-order. In the table, the columns containing second-order word associations are labeled 2nd word.
Through this network, CORVID automatically delineates the relationships between concepts as they are associated in the natural language sample being analyzed. Connecting together words across orders in the CORVID feature creates aptly labeled revelatory statements, which both reveal and represent popular means of relating concepts in the natural language sample (and, by extension, the words and thoughts of the person or people supplying it). These revelatory statements, therefore, relate the main concepts of your natural language sample in words and ways that maintain high degrees of fidelity to their original expression.
4.4 Bracket accordingly.
Once you have constructed revelatory statements for each of the main concepts in each part of your natural language data set, bracket to ensure that there are no unaccounted geists or other variables potentially influencing your structure. See step 2 for the specific procedures related to bracketing for geists and other influential variables.
4.5 Reiterate, as needed.
If the process of bracketing appears to reveal additional functional associations between your concepts, repeat substeps 4.1-4.4 until they are generally accounted for in your statements.
If, after you have considered the associations between your concepts, you have not changed the original structure of the data in your spreadsheet, you may proceed. If, however, you repartitioned it based on the considerations in this step, you will need to upload it to Raven’s Eye as a separate project, categorize your columns, and select it for analysis2. By default, you will arrive at the main table of your newly uploaded project. If you have substantially revised your concepts, it is advisable to revisit step 2 and proceed from that point in the procedures.
Many researchers may find the needs of their projects met upon completion of this procedure. If such is the case in your project, you can download the results you have been otherwise visually inspecting, and produce messages, tables, graphs, or other visualizations in your spreadsheet, word processing, or other programs.
If, after you have considered the associations between your concepts, you have not changed the original structure of the data in your spreadsheet, you may proceed. If, however, you repartitioned it based on the considerations in this step, you will need to upload it to Raven’s Eye as a separate project, categorize your columns, and select it for analysis2. By default, you will arrive at the main table of your newly uploaded project. If you have substantially revised your concepts, it is advisable to revisit step 2 and proceed from that point in the procedures.
Many researchers may find the needs of their projects met upon completion of this procedure. If such is the case in your project, you can download the results you have been otherwise visually inspecting, and produce messages, tables, graphs, or other visualizations in your spreadsheet, word processing, or other programs.
Notes.
1 To be precise, Giorgi (2012) refers to this process as, “free imaginative variation,” (pp. 5-6). In this process, one utilizes the statements present in the raw data to imagine the many and various ways that the same or similar information could be validly expressed by others during other valid instances of the same or similar experience. While doing so, one pays attention to the affect that altering specific aspects of a given statement has on its underlying meaning. Those alterations that facilitate an explicit and direct revelation of the meaning in the experience are maintained, while those that are not are set aside for the time-being. Performing this process iteratively helps to distinguish those aspects of a given experience that are apparently essential to it, from those that are not.
2 In this way, your major data transformations are demarcated by file, and thus facilitate easier replication and extension in subsequent projects.
2 In this way, your major data transformations are demarcated by file, and thus facilitate easier replication and extension in subsequent projects.
Reference.
Giorgi, A. (2012). The descriptive phenomenological psychological method. Journal of Phenomenological Psychology, 43, 3-12. doi: 10.1163/156916212X632934