On the surface MAXQDA and ATLAS.ti seem almost identical. Yet when we look under the hood, we see strong differences: one follows the logic of a relational data base and sorts everything into neat categories and the other operates like a graph data base that links different entities to form a large network. The implicit potentials and constrains of each (and any) software commonly drive our research because we too often follow the road that we already know best.
Since my master thesis (Müller 2017), I am using MAXQDA. It’s core function is to structure text by segmenting it with labels. However, I never noticed, how MAXQDA structured my research too. As a freelancing software trainer, I was time and again asked to offer ATLAS.ti as well, so I moved on from a software I already knew well, to another one that looked very similar.
To familiarize myself with ATLAS.ti, I rebuilt the data base of my master thesis within it. Yet strangely enough, this proved difficult to do and seemed not to make much sense. Soon, I noticed major differences that had strong methodological implications.
Commonalities: Twins on the Surface
ATLAS.ti and MAXQDA are both around since the 1990s and quite prominent in qualitative sociology. They belong to the group of QDAS, qualitative data analysis software, a type of software that is primarily used to structure text by manually assigning labels to individual segments (also images, audio or video are possible). By this, they bring together segments of different documents and/or data types under one keyword (usually called: code). Furthermore, they count the segments and offer many options to visualize and explore the structured data.
Differences: Table or Network logic
Although both basically use SQL-databases, there is a striking difference in their internal logic. While MAXQDA follows the structure of a relational data basis, ATLAS.ti is organized like a graph data base and this strongly influences the way each software is used.
MAXQDA works like a classification system. Everything is sorted into well defined categories. For each document individual variables can be defined. These document variables allow for close cooperation with many statistical programs and follow exactly the same logic. MAXQDA even features its own integrated statistics modul that works as a slim and light version of SPSS (Compare: Rädiker, Kuckartz 2019, 129-131)
In sorting everything into well defined, hierarchical categories, the program joins forces with the method “Qualitative Content Analysis” (Mayring 2015, Kuckartz 2014). The coded segments are sorted into a tree structure built of several layers of codes and sub-codes. The analysis is usually done with a Mixed Methods approach that strongly incorporates qualitative and quantitative methods.
ATLAS.ti on the contrary neither uses code hierarchies, nor variables. In its core, ATLAS.ti is a system of nodes and relations. Each segment that is created becomes a node in a network. “Coding” a segment here, is not putting it “into a category” but rather linking a text segment to a label (compare: Friese 2019, 192-194).
Further on, researchers can create semantic linkages between the different elements. A text segment can be formally defined as “contradicting” another text segment. A code can be defined as “being part of” another code. In this ATLAS.ti creates a system of linkages between different entities (documents, codes, segments, memos etc.). In such a system “document variables” make little sense. However, “document groups” or “code groups” can be used as quasi-dichotomous variables for a similar purpose. For example, a document group “gender: female” can be linked to a document “Interview with Mrs P.” to identify the interviewee as a woman.
In this, ATLAS.ti joins forces with a method called “Grounded Theory” (e.g. Strauss and Corbin 2010) that works quite contrary to Qualitative Content Analysis. Where Content Analysis usually creates a proper set of well defined categories and applies it to the text, Grounded Theory creates a vast abundance of codes during “Open Coding” and after that begins to link, merge and relate the codes to each other (called “Axial Coding”). Therefore, it is set out to find patterns within a text, instead of applying a well trimmed analysis tool.
Implications: Blind Spots and Focus
Of course, both programs are not as limited as it may seem. Both can be used “inductively” as well as”deductively”. Some linkages are possible in MAXQDA and one can recreate hierarchies by using relations in ATLAS.ti, if they are needed. However, the suggestive power of each software makes this unlikely. Instead, the tools strongly reflect and I would argue even reproduce the methodology that led to their creation.
When I started to replicate my thesis in ATLAS.ti (as mentioned above), I ran into the problem of translating all my document variables (gender, place, date, occupation, sentence, execution date….) into this format. On the other hand, as soon as I called up my text segments in the “Network” view, I could right away start drawing explicit semantic linkages between the different text segments.
Inspired by this encounter with ATLAS.ti, I came up with a small research project to model statements pro and contra the existence of witchcraft within an 18th century encyclopedia. With ATLAS.ti I could easily show how both sides were using the same sources for their arguments (for example passages from the bible) but drew completely different conclusions from them. This way, I could quickly construct a discursive network that would show me the underlying logic and conflicts of the argument.
“Which software is better?” would not be a meaningful question. Each software was built with a different purpose in mind. I enjoy working with the semantic networks of ATLAS.ti, yet the Mixed Methods tools within MAXQDA are also very powerful. For me, both will remain, not just in my teaching portfolio, but also in my toolbox for future research.
Beyond this comparison, whatever tool we use, we must not forget that it is never methodologically neutral. The moment we come up with a plan, we think in the logic of tools. If we are familiar with graph data bases, all that we see will be nodes, linkages and networks. If we are used to relational data bases, we will only see cases, variables and tables.
Usually, the research question should drive our research, yet instead, we are often driven by the tools and the methods with which we are already familiar. Therefore, it may proof valuable to once in a while go back to the tool store and see what else might be in stock.
Update 01/12/2019: With the implementation of the REFI-QDA project exchange standard it has become possible to move projects from ATLAS.ti to MAXQDA and vice versa. You can read more about this exchange format in this article.
4 thoughts on “MAXQDA or ATLAS.ti? How software shapes research”
I do not agree with the basic premise of this article. You should never allow a software to drive your research. It is s matter of knowing the methodology that one wants to use and the tool (see Woolf and Silver, 5 level QDA). The categories from MAXQDA with sub codes can easily be transferred to ATLAS.ti using prefixes for the sub codes – if your aim is to develop a hierarchical code system. I always strongly advise against using the network function to organise codes using relations. This will inhibit you later when linking codes conceptually. See Friese, 2019 :-).
Regarding the work with variables, one just needs to know that one needs to use document groups. It is also possible to import variables from an Excel spreadsheet where variables are listed in the usual way.
The code document table and code co-occurrence table give you numbers (plus the data behind it), absolute frequencies, relative frequencies, you can normalise and binarize data ; export all codings as SPss syntax file for further statistical analysis, or as Excel file for import in R, SAS or other statistical packages.
There are of course differences between the packages, some of it can be traced to the Origins if the software and are visible in the software architecture (which then allows for doesn‘t allow to do certain things, or make it a bit cumbersome). But I object the view, that a particular piece of software drives your analysis. The person using it should be in charge 😉
Thank you very much for your critical comment and references for further reading! In short, I can only say: I agree with you. Yes: “The person using it should be in charge.” (Friese 06/09/2019). However, I think, this is often not the case.
As an early career researcher, it was an eye opener for me to start working with ATLAS.ti as I noticed only then, how strongly the “MAXQDA-methodology” was influencing my work. I think the crucial steps to overcome this blindness are knowing more than one software, more than one methodology and knowing at least one methodology very well. I am working on all three of them 🙂
Thank you so much for this information.
I want to analysis my data that is only transcripts of interviews for a qualitative study. For this qualitative study, I am focusing on (1) students’ perceptions of project-based learning experience in high school and (2) how are project-based learning experiences influential to students’ college-readiness?
I was wondering in this qualitative study when there is NO documents groups in Atlas ti, should I use the “Network” feature of the software for the analysis? Especially, when considering students’ experiences as a phenomenon for the study. I would appreciate your reply.