Why I Keep Rereading the Same 20 Papers

My desk has a stack of papers held together by a binder clip. The top page of Whetten (1989) has three different colors of marginalia. Blue ink from the first reading last semester. Pencil from the second. Red pen from the third. The date stamps I started writing on the corner of each paper show I last read Sutton and Staw (1995) on February 14, then again on March 28, then again on April 30. I did not plan to read these papers four times. It just kept happening because each time I came back, the paper was different.

The first time I read Whetten (1989), I understood the four elements. What, how, why, boundary conditions. I wrote that down, nodded, and moved on. The second time, I started seeing what Whetten was actually arguing against. He was not just defining theory. He was responding to an academy that was accepting papers with no mechanism at all. The third time, I noticed something I had skipped entirely on the first two passes: the seven reviewer questions at the end. What is new? So what? Why believe you? Who cares? Those questions changed how I read every paper after that, not just Whetten. The fourth time, I connected the specificity of those questions to Sutton and Staw (1995), which names the five things that are not theory: references, data, variables, diagrams, hypotheses. The two papers are two sides of the same coin. Whetten builds the positive definition. Sutton and Staw build the negative definition. Reading them together, after reading them separately across four sessions, made them lock into place in a way that reading them once could never have done.

This pattern keeps repeating with a set of maybe twenty papers. Markus and Robey (1988) I read three times before the emergent perspective became something I could apply rather than just recite. The first pass I got the three causal stances: technological imperative, organizational imperative, emergent. The second pass I got what really matters, which is that the emergent view is almost always the correct IS answer but most published research defaults to one of the other two. The third pass I started seeing it everywhere. I wrote about this separately on this blog, because it turned into its own obsession. Orlikowski and Iacono (2001) took me two passes to even understand what the five views of the IT artifact meant and a third to grasp the force of that 88 percent statistic: eighty-eight percent of ISR papers they reviewed did not use the ensemble view. The artifact was present in name but absent in theory. Benbasat and Zmud (2003) diagnosed the same problem from a different angle, arguing the discipline had drifted from its core. Sarker et al. (2019) confirmed it empirically eighteen years later: fifty-six percent of IS papers are Type I on the sociotechnical axis, meaning you can remove IT from the model and nothing changes. Reading those three across time, rather than in sequence, let me see the argument build across two decades. Each paper cites the earlier ones. Each one refines the diagnosis. The pattern would be invisible if I had read them once and moved on.

I started keeping a log. Every time I sit down to reread one of these papers, I write the date and what I am looking for. The first entry for DiMaggio and Powell (1983) says "understand the three isomorphism types." The second entry says "find the mechanism that drives mimetic isomorphism." The third entry says "trace how institutional theory connects to TOE and structuration." The fourth entry says "figure out why coercive pressure is external, not internal." That last one came from an oral exam trap I had flagged on oral.html: practitioners keep calling a CEO mandate coercive, but in DiMaggio and Powell the categories are clear, coercive comes from external regulation like GDPR or FedRAMP, not from the boss. I would never have caught that distinction on a first read. I only caught it because I was reading the paper a fourth time with specific exam traps in mind.

Gregor (2006) is the one paper in the stack that I think I finally understand after five reads. The five theory types seemed simple the first time. Type I is analysis. Type II is explanation. Type III is prediction. Type IV is explanation and prediction. Type V is design and action. The trap I kept falling into was treating Type IV as the gold standard for everything, which misses the point that each type serves a different purpose and is evaluated by different criteria. A Type I taxonomy like Sarker et al. (2019) does not fail at prediction, that is not what it is for. Gregor also discusses Dubins precision paradox and power paradox, which I completely missed on the first two reads because I was too busy memorizing the five types. A precise theory loses generality. A general theory loses predictive sharpness. You cannot have both. That tradeoff is the kind of thing you only notice when you are not reading to pass a quiz but reading to build an argument.

I think there is a difference between broad reading and deep reading, and comps prep made me learn it the hard way. Broad reading gives you coverage. You know what every paper says, roughly. You can name the authors and the years. Deep reading gives you the ability to use a paper as a tool. You know not just what Whetten claims but where the claim breaks down. You know that Sutton and Staw are useful for the negative definition of theory but that their five non-elements have been debated and refined by later scholars. You know that Markus and Robeys emergent perspective is not a compromise between the other two views but a fundamentally different logic about causality. That kind of knowledge only comes from returning to the same page months apart and finding something you missed.

I have a theory map growing on the wall next to my desk. I wrote about it on this blog a few weeks ago. Every time I add a new connection between papers, I go back and reread both papers to make sure the connection is real and not something I am imagining. The connection between Baird and Maruping (2021) and Burton-Jones et al. (2021) took me three passes to confirm. Baird and Maruping argue that delegation replaces use as the central construct for agentic IS. Burton-Jones et al. propose four theorizing strategies: replace, reformulate, extend, envision. Baird and Maruping are doing reformulate, not extend. They are not adding a moderator to TAM. They are changing the foundational assumption from human as sole actor to AI as co-actor. I could not see that distinction until I had read both papers closely enough to understand what each was really doing with theory rather than just what each claimed about technology.

The twenty papers are not the only ones I will have read by the time comps are done. There are one hundred eighty four papers in the study hub. But the twenty are the load bearing walls. Whetten (1989) and Sutton and Staw (1995) define what theory is and is not. Markus and Robey (1988) define how to think about causality in IS. Orlikowski and Iacono (2001), Benbasat and Zmud (2003), and Sarker et al. (2019) define the identity of the field. Gregor (2006) gives the typology of theory purposes. DiMaggio and Powell (1983) anchor institutional theory. Mohr (1982) separates variance from process logic. Burton-Jones et al. (2021) and Baird and Maruping (2021) show where the field is going. Every time I read one of these again, I find a corner of the argument I had flattened on earlier passes. The binder clip is getting worn. The marginalia is getting dense. I think that is what readiness actually looks like.