ATLAS.ti is intended primarily for supporting qualitative reasoning processes. On the other hand, especially with large amounts of data, it is sometimes useful to analyze the data using statistical approaches. ATLAS.ti can export your data in form of a syntax file for SPSS®, and a generic Excel format that can be imported into software like R, SAS, STATA as well as SPSS.
The basic components for statistics are cases and variables. The statistic export function in ATLAS.ti treats codes as variables and data segments (quotations) as cases.
The notion of a "case" here is rather fine-grained and differs from the common understanding of this term. Usually cases in qualitative research refer to persons, interviews, or documents. We chose to treat the smallest unit as a case for the statistical export, to ensure that no data is lost. Broader information, e.g., which document or document group a quotation belongs to, is coded into the various variables.
In contrast to the dichotomous treatment of codes within ATLAS.ti, you can use codes for further statistical analysis as ordinal or interval scaled variables by using a specific code-naming convention.
Within ATLAS.ti, a code is always dichotomous, because it either refers to a given quotation ("1"), or it does not ("0"). Each case (= quotation) can, in respect to the codes, be described as a vector of 0's and 1s. The concept of scaled codes/variables requires a special syntax.
If you for instance want to code for three different levels of evaluating something: very good, good, neither god nor bad, bad, very bad, A special naming convention (name%value) is necessary to let ATLAS.ti identify scaled codes from dichotomous codes. When coding the data, the code category needs to look as follows:
- evaluation%1 very bad
- evaluation%2 bad
- evaluation%3 neither good nor bad
- evaluation%4 good
- evaluation%5 very good
This notation allows the system to construct one variable from all codes with the same prefix. The variable name will be "evaluation", and the values for this variable are:
- 1 very bad
- 2 bad
- 3 neither good nor bad
- 4 good
- 5 very good
Instead of the % as separator, you can also choose a different separator. If so, you need to indicate this when preparing the SPSS syntax file.
You can use string or numerical values; anything that follows the special symbol is interpreted as a value.
Keep in mind that ordinal codes only have meaning in the context of the statistic program you are using. Within ATLAS.ti, the differently valued codes are treated like every other code - as 0 or 1: has been applied, or has not been applied.
Do not assign more than one scaled variable value to the same quotation. Although ATLAS.ti permits an arbitrary number of codes to be attached to a quotation, this would not make much sense with mutually exclusive values of scaled variables. If you do so, the SPSS generator will simply ignore additional values after processing the first one it finds for a given quotation. Since it cannot be guaranteed which value will be detected first, this will most likely produce unpredictable results.