Hypotheses and Competing Explanations

Introduction

The first explanation that fits a result is rarely the only one. A battery loses capacity; possibilities include consumed active lithium, isolated electrode particles and increasing polarization. A reaction accelerates after a powder is added; the powder may catalyze chemistry, change pH, adsorb an inhibitor or alter light scattering. Research becomes convincing when the experiment is designed to distinguish these explanations rather than merely collect more observations that all of them predict.

Core explanation

A hypothesis is a proposed causal account, not a synonym for a guess. It should yield predicted observations under specified conditions. List at least two alternatives and a baseline. Then ask, “What measurement would give different expected results under these accounts?” If all hypotheses predict the same observed trend, that trend has little discriminating power even if measured very precisely. A strong experiment targets disagreement between predictions.

Consider a colored solution that becomes pale after adding a solid. Hypothesis A says the dye chemically decomposes; hypothesis B says it adsorbs on the solid; hypothesis C says the solid scatters light and alters the spectrometer reading. A simple absorbance decrease is compatible with all three. Separate the solid and measure dye in the liquid and on the solid, identify products with chromatography or mass spectrometry, and measure a particle-only optical blank. A mass balance and product identification distinguish chemistry from transfer or optical artifact.

Mechanistic tests often use interventions. Changing temperature can reveal activation behavior, but different transport or equilibrium effects may also be temperature-dependent. Isotope labeling can trace atom movement, provided labeling does not itself change the mechanism substantially. Altering a reactant concentration can reveal reaction order under specified conditions, though apparent order may change with adsorption or mass transport. A battery's rate dependence can separate equilibrium capacity from kinetic access, while open-circuit relaxation helps detect polarization. Each method has alternative interpretations, so predictions and controls should be written explicitly.

Do not confuse consistency with exclusivity. A spectrum matching a proposed intermediate supports its possible presence, but another species may share a feature and the intermediate may be a spectator. Temporal coincidence between a peak and product formation is stronger if the peak grows and disappears in a predicted sequence, yet still may not prove it is on the productive pathway. Selective perturbation—changing the concentration of a suspected intermediate while holding other factors fixed—can add causal leverage.

A null model should be concrete. “Nothing happens” is not always the right baseline. For a catalyst, uncatalyzed background reaction may occur; for an analytical signal, a nonzero blank may be normal. Quantify baseline behavior and its variation. Make predictions before seeing the new data, ideally with ranges rather than vague “increase” or “decrease.” If the observed data contradict both favored and competing mechanisms, revise the model instead of forcing a fit.

The NIST experimental statistics handbook emphasizes design and analysis aligned with the intended comparison. Mechanistic discrimination adds a chemical requirement: chosen variables must have different predicted consequences under plausible chemistry. More replicates improve precision, but they do not rescue a measurement that is non-discriminating by design.

Step-by-step reasoning

Write the observation without interpretation. Construct a short list of plausible chemical and measurement explanations. For each, predict what should happen to at least two independent observables under a controlled perturbation. Choose a feasible test where predictions diverge most. Add blanks, references and independent repeats. Analyze whether the result excludes, weakens or merely remains compatible with each hypothesis. Record unresolved alternatives honestly.

Visual explanation

Draw three columns labeled decomposition, adsorption and optical scattering. Rows are tests: dye found on solid, new molecular products, and particle-only blank. Mark predicted high or low results in each cell. A useful test has different marks across columns; a row where every column predicts low solution absorbance is not useful for discrimination. This matrix can guide which measurements are worth performing first.

Real-world analogy

A wet floor could result from a leaking pipe, an open window during rain or a spilled drink. Seeing water proves the floor is wet but does not distinguish causes. Checking the pipe, weather and nearby glass can separate them. Chemical research follows the same logic, though the “clues” may be spectra or isotope patterns rather than visible objects.

Real-world example

A solar cell's fill factor falls after heat exposure. One group attributes it to a changed absorber band gap. Another suggests contact resistance. Measure optical absorption to test a large gap change, then compare series-resistance-sensitive current–voltage behavior and contact imaging. If absorption is stable but contact resistance grows and delamination appears, the contact hypothesis gains support. It remains possible that subtle absorber defects also contribute, so the final claim should match the resolution of the tests.

Why?

Why collect a second type of data rather than repeating the same curve many times? Repetition narrows uncertainty around that curve, but multiple mechanisms may still generate it. An independent measurement aimed at a predicted chemical or structural difference can shrink mechanistic ambiguity. The goal is not simply more data points; it is information that changes the relative plausibility of competing explanations.

Common misconception

“A hypothesis is proven when one prediction comes true.” Many hypotheses can predict the same outcome, and a prediction may be too broad to be informative. “The simplest explanation must be correct” is a useful preference only when alternatives fit comparably and assumptions are stated. “A mechanistic cartoon is evidence” reverses the logic: the cartoon summarizes a hypothesis that experiments must test.

Worked example

A battery's capacity falls from 2.0 to 1.6 Ah at 2 A. At 0.2 A, the aged cell still delivers 1.9 Ah. If 0.4 Ah of active lithium were permanently lost, the low-rate capacity would normally also be greatly reduced under the same voltage window. The strong rate dependence instead points toward polarization or transport as a major contributor, though the 0.1 Ah low-rate loss may still involve inventory or active-material change. Compare impedance and voltage relaxation with a fresh control, then assess whether electrode structure or contacts changed. The two-rate experiment discriminates better than another 2 A measurement alone.

Quick check

1. Why does a single loss of absorbance not prove that a dye has chemically decomposed? Answer: Adsorption onto a solid and optical scattering can also lower the measured solution absorbance. Product identification, mass balance and appropriate optical blanks can distinguish these explanations.

Exam focus

Name the observation separately from the proposed cause. Give at least two plausible mechanisms and a test with different predicted outcomes. State controls and what each possible result would imply. Do not claim that a method excludes a mechanism unless its sensitivity and sampling are sufficient. Use cautious terms such as “supports” when alternatives remain.

Advanced insight

Information gain provides a formal way to choose experiments: favor measurements whose predicted distributions differ strongly between hypotheses relative to their uncertainty. This can justify a costly, targeted measurement over many cheap repeats of an ambiguous one. Predictions may come from kinetic models, simulations or prior experiments, but they should be updated when data reveal unmodeled chemistry.

Summary

Mechanistic research compares explanations that can all fit an initial observation. A discriminating experiment seeks a controlled condition or measurement where those explanations predict different results. Blanks, quantitative baselines and independent methods protect against measurement artifacts. Evidence can support a mechanism without uniquely proving it; good reporting states the remaining alternatives.

Practice questions

1. A catalyst accelerates conversion but also raises solution pH. Suggest a discriminating control. Answer: Run a catalyst-free reaction adjusted to the same pH and compare rates, while holding concentrations, temperature and mixing fixed. If pH alone reproduces the acceleration, direct catalysis is not uniquely established.

2. A transient spectrum appears during reaction. What additional observation would strengthen a claim that the species is on the productive pathway? Answer: Show that its formation and decay follow a predicted kinetic sequence and that selectively changing its availability alters product formation as predicted, while controlling for side effects of the perturbation.

3. Both proposed battery-degradation mechanisms predict lower capacity at high rate. What is wrong with only measuring another high-rate capacity curve? Answer: The measurement is not discriminating. A low-rate curve, impedance, relaxation or inventory analysis that predicts different outcomes under the mechanisms is needed.