As I reached the midpoint of my research, I ran into a problem that could not be fixed by simply finding a better data set or reading more papers. The project itself had stopped making sense in the form that I had originally planned. In the beginning of the summer, I had hoped to create a simple field-based tool that would connect visible forest habitat features with different aspects of underground ecological recovery. The basic logic behind this idea had originally seemed strong as fallen deadwood, plant diversity, litter and canopy cover all influence the environment where soil microbial communities live. My assumption had been that the most difficult part would be identifying the right numerical thresholds, and combining them to create a useful and practical index, but I had underestimated the complexities of this project.

By around weeks three and four of my research, I had amassed a large body of literature, and also got into the habit of going back through the most important studies and checking the full papers rather than relying on abstracts or citations from other sources. As I did a more thorough comb-through, however, I found that the literature supported the ecological relationships behind the project more than the numerical system itself that I was attempting to build. To me this first seemed like a literature-search problem. I kept looking for papers that would back up my ideas, finding new ways to specify my search. I followed citations backwards, searched for papers that used similar measurements, broadened my search from New England to other temperate forests, checked the papers against their primary sources, and tried to obtain full texts when abstracts were simply not enough. The more carefully I looked, however, the clearer it became that the missing evidence was not necessarily hiding somewhere waiting to be found. In some cases, the studies were answering a completely different question.

One good example of this can be seen in deadwood. There was substantial evidence that fallen wood matters ecologically, but my field measurement was not “is there a log next to this soil sample.” I was looking at total coarse woody debris volume across a forest plot, and although they may be related measurements, they are not interchangeable. The same problem appeared with litter. Experimental studies show that the presence of litter affects fungi, microbial activity, or decomposition. But measuring the standing depth of litter at one point in time is much different from manipulating the litter layer experimentally. 

Then there were the numerical thresholds themselves. I had expected to find evidence supporting specific points where a score below the threshold would indicate one condition and a score above it would indicate the other. Instead, I found either continuous relationships, context-dependent relationships, nonlinear responses, or studies that claimed ecological importance without providing any threshold at all. By this point I realized that I had not failed to locate the right number, but that I was wrong in believing such a number could exist. 

It was frustrating seeing my original research plan somewhat falling apart as I had already spent a lot of time developing the original structure. Part of me wanted to keep going and force my original plan to work, but instead I started reviewing all of the assumptions behind my project one by one.

I separated two questions that I had previously allowed to be lumped into one:

• Is this habitat feature ecologically important?

• Does this specific field measurement provide a defensible numerical proxy for the ecological outcome I want to interpret?

The distinction between these two questions became my main troubleshooting method. Instead of asking whether a paper “supported” deadwood, litter, or plant diversity in general, I began writing down exactly what it supported. For some, it was evidence for biological importance. For other studies, it was evidence for the actual field measurements, evidence for numerical thresholds, or evidence showing that the relationship was conditional or contradictory. 

I also had to take measurement definitions much more seriously. I quickly discovered that even something as seemingly simple as “litter depth” could refer to different physical layers depending on the study. Woody richness could change depending on the size of the lot and the minimum stem diameter. Deadwood estimates depended on the minimum diameter included as well as the sampling method. 

I started spending more time reading the methods sections than introductions and abstracts. At the same time, however, I was trying to manage the problem of looking at too much research all at once. This project had expanded into dozens of studies and multiple datasets, source lists, evidence tables, model versions, and field-protocol decisions. At this midway point my head was constantly full of questions such as whether or not a certain variable was actually comparable, if I had already checked a source or not, or whether the paper was even relevant. 

The best way I found to manage this mess in my head was to make the process more explicit. I began keeping clearer records of what each source actually established, and separating calibration evidence from ecological evidence. I also became more disciplined about stopping my searches when I found that I was continuing just because I disliked the answer. This was probably one of the harder lessons I learned at this midpoint, the fact that there is always another search term, citation or dataset. It can be difficult to know where more work meant increased confidence, and when it was just simply increasing volume.

This pivot gave me a new, fresh way to view my project. I didn’t necessarily need to assign a microbial meaning to every forest measurement in order for those measurements to be useful. Even without the microbial context, a land steward measuring deadwood, native woody richness or litter depth still faces a basic interpretive problem: is the amount they measured relatively low, typical, or high? Existing forest datasets could answer this question much more directly than the literature could support a microbial-recovery score. This idea became the beginning of my project’s redesign. Instead of asking “What microbial score should this measurement receive?” I began asking, “Where does this measurement sit within an appropriate forest reference distribution?” Although it may seem like a small change now, at that moment it meant letting go of the central idea I had previously come up with and was expecting to work on for the rest of the summer. 

Midway through this project I was not completely sure what the final product would look like, but I did know that I was not going to keep an old model just because I had already invested some time in it. The solution was not to make the evidence fit, but to change the project itself.