SUPPORTING GENERAL HYPOTHESES

Frameworks can contribute to supporting General Hypotheses.

If data are consistent with the predictions of Measurable Hypotheses, the data can be considered to support the General Hypotheses that led to the predictions of the Measurable Hypotheses. 

DEFINITION: "Support" for a hypothesis has a specific meaning: the data of the current experiment did not reject the predictions of the hypothesis. 

However, simply failing to reject a particular General Hypothesis of a study is only one piece of evidence, and may not alone be sufficient reason to continue research to test and further develop the General Hypothesis. Therefore, one role of the Discussion can be to provide additional support for General Hypotheses.

Additional support for General Hypotheses can involve

1) Defending the assumptions used in the reasoning of the study.

2) Explaining how the findings of the study and the General Hypotheses are consistent with broader scientific understanding.

1) Defending the assumptions used in the reasoning of the study.

Similar to a forthright discussion of experimental limitations, identifying the major assumptions of the study can help establish the reader's trust. Moreover, clearly identifying assumptions can anticipate probable questions, and prevent unanswered questions from undermining arguments about the General Hypotheses. Therefore, it can be helpful to provide readers with a clear explanation of each major known, assumption made in designing and conducting the study.

The assumptions that could potentially affect research vary considerably by field. Examples of assumptions in research involving humans could include assuming that sex or gender of study participants does not affect physiology or performance, an assumption that convenience samples (often college students for university research) represent a broader population, assumptions that important variables do not substantially change with age, assumptions that behavior in laboratory settings transfers to behavior outside the laboratory, etc. Although animal models are critical for biomedical research, much research on animals assumes that principles learned from animals also have relevance to humans at the molecular, physiological, or even behavioral levels.

Assumptions are not limited to biology. For example, the physical sciences and engineering commonly study systems that can be "linearized:" investigated in narrow ranges where the responses of systems are linearly related to inputs. Principles like the "Ideal Gas Law" assume that simplified relationships apply broadly to many different compounds. Clearly, researchers make assumptions in almost every field of science.

There is no shame in making assumptions. However, if authors do not recognize and address important assumptions, readers can be confused, withhold judgment, not agree with the arguments of the study, or lose trust in the competence of the authors (or all at once). Simply avoiding mention of important assumptions is not a viable strategy: competent scientists will be able to "read between the lines," and do not appreciate subterfuge. Therefore, it is in the authors' best interest to voluntarily identify the major assumptions of a study.

Similar to the limitations, a useful framework for explaining assumptions is:

1) Identify the assumption made, and why the assumption was necessary.

2) Explain using a reasoned argument why the assumption does NOT affect the conclusions of the study (e.g. the tests of the Measurable Hypotheses in the Results).

Many students perform step (1) and identify assumptions without performing step (2) and explaining why the assumptions do NOT affect the conclusions! Readers are therefore forced to come to conclusions on their own (and scientific readers are not inclined to be charitable, particularly when expected to do work for the authors). Therefore, it is critical to perform step (2) and make a clear, evidence-based argument why an assumption is NOT likely to affect the conclusions of the study. 

Addressing the assumptions can involve references to other studies, alternative analyses of data or limited additional calculations as necessary. For example, the assumption that sex differences do not affect performance could be supported by the results of similar studies that tested for (and did not find) sex differences.

2) Explaining how the findings of the study and the General Hypotheses are consistent with broader scientific understanding.

One framework that can help to organize arguments to support General Hypotheses is inductive reasoning using Hill's Criteria. 

Examples of how Hill's Criteria could apply to the Discussion include:

1) Reliability – Do repeated studies all lead to the same conclusions?

Do the data collected by the present study match data collected in previous studies? Finding that the data are quantitatively consistent with other research can strengthen confidence in the Methods of the study, the resulting data and conclusions of the Results, and also contribute to supporting shared General Hypotheses. An example of an argument for reliability could involve comparing the results of complex calculations of arm movement to previous measurements: "The elbow excursions of 77 ± 11° that the monkeys used for the present task were comparable to the 81 ± 20° excursions reported by Christel and Billard (2002)" (Jindrich et al., 2011).

2) Diversity – Does evidence from many different approaches all support the hypothesis?

Do different types of studies all support the same General Hypothesis? If a diversity of approaches are all consistent with a hypothesized explanation, then the explanation is more likely to be a general, valid explanation. The Discussion can make arguments for diversity by surveying a wide range of literature and finding consistent support for a General Hypothesis. For example, "Similar differences between ‘‘massed’’ and ‘‘distributed’’ practice were observed in motor learning paradigms other than adaptation (Lee and Genovese 1988), as well as in verbal learning paradigms (Ebbinghaus 1885; Glenberg 1979)" (Bock et al., 2005), or "That exercise was equally effective [in reducing symptoms of depression] as medication after 16 weeks of treatment is consistent with findings of other studies of exercise training in younger depressed adults [14,15,17,18]." (Blumenthal et al., 1999).

However, as always, it is important to make sure that arguments in the Discussion are a valid representation of the research on a topic. Arguments for Diversity should not represent "cherry picking" in the service of confirmation bias.

3) Plausibility – Are there reasonable mechanisms that underlie observed outcomes? Are the mechanisms consistent with, and do not conflict with, other knowledge?

Consistency, or "consilience," of scientific explanations is extremely important for science. For example, proposed biological mechanisms must be consistent with known laws of physics and chemistry (e.g. conservation of energy, entropy, etc.). Physiological or behavioral explanations must be consistent with known physiological or neural processes. Therefore, plausibility is an important and common argument in the Discussion.

Two approaches to arguments for plausibility are (A) information from other studies suggest reasonable mechanisms to explain data observed in the current study; or (B) data from the current study provides direct mechanistic evidence for General Hypotheses. An example of the first type of argument is: "Animal research suggests that [differences between ‘‘massed’’ and ‘‘distributed’’ practice] may be related to differential modulation of protein synthesis- dependent molecular processes which affect the expression of synaptic connectivity (Genoux et al. 2002; Scharf et al. 2002)" (Bock et al., 2005).

4) Experimental Interventions – Can direct interventions produce predicted outcomes?

Sometimes General Hypotheses are developed from first principles, physical models, or observed correlations. Direct experimental testing of General Hypotheses is an indispensable tool for science. The Discussion can include arguments that experimental data supports scientific explanations or models. For example, "Our results suggest that humans show body control strategies that result in relationships among movement parameters that are consistent with the distributed feedback rules used by Raibert’s robots" (Qiao and Jindrich, 2012).

5) Temporality – Are there time-based dependencies (e.g. causes precede effects)?

Time-based arguments are particularly important for hypotheses that involve causal relationships. Effects are commonly observed after causal phenomena. The Discussion can include time-based arguments to support hypotheses. For example, "It is clear that neuronal processes that precede a self-initiated voluntary action, as reflected in the readiness-potential, generally begin substantially before the reported appearance of conscious intention to perform that specific act" (Libet et al., 1983).

6) Strength – Is there a strong association between variables? 

Although statistical tests can test for differences among groups, statistical tests alone do not address whether differences among groups are important. Demonstrating that there are strong associations among variables can be an important part of arguing that statistically-observed differences are important. The Discussion can compare findings to other phenomena to make an argument that observed relationships among variables are strong and important. For example, "The magnitude of reductions in depression scores is also compatible to the levels achieved using sertraline in other clinical trials of depression [45,48]. Moreover, the changes in depressive symptoms found for all treatments in our study are consistent with the extent of improvements reported in more than a dozen studies of psychosocial interventions for MDD [12,49-53]" (Blumenthal et al., 1999).

7) Specificity – Are there specific factors (i.e. not all factors) that result in observed outcomes?

Specificity can be important for using Strong Inference to reject alternative hypotheses. If General Hypotheses lead to specific predictions that are consistent with data (whereas the predictions of other hypotheses are not), the General Hypothesis may be stronger than alternatives. For example "It became obvious that the improved stepping associated with step training occurred as a result of the repetitive activity of those spinal locomotor circuits that actually generated the load-bearing stepping, since spinal cats that were trained to stand bilaterally learned to stand but could not step as well as even those spinal cats that were not trained at all" (Edgerton and Roy, 2009).

8) Biological gradient – Are there biological gradients or dose-response relationships?

Experimental studies may directly test for dose-response relationships. For example, "Quipazine increased the sensitivity of the spinal cord to ES. The stimulation threshold to elicit muscle twitch as detected visually and by palpation was lower after quipazine administration (Table 1)... There was a significant decrease in the effective ES intensity after administration of quipazine at dosages of 0.2, 0.3, and 0.5 mg/kg (Table 1)" (Ichiyama et al., 2008).

Even if an experimental study does not directly test for biological gradients, using the results of similar studies can allow for experimental data to contribute to arguments for a biological gradient. 

The Discussion can focus on making a limited number of strong arguments.

Experimental studies do not need to include all research findings and hypotheses in a particular field in the arguments of the Discussion. For example, a study based on a single experiment does not need to address all of Hill's Criteria. Other types of scientific papers (e.g. reviews and meta-analyses) are available to make comprehensive arguments about important research topics that survey the entire literature. Instead of trying to address too many topics in the Discussion, focusing on fewer, stronger arguments can be sufficient to make a positive contribution to scientific understanding.

Papers can typically devote 3-5 paragraphs of the Discussion to supporting General Hypotheses. Three to five paragraphs may not allow strong arguments based on all of Hill's Criteria. Therefore, it can be acceptable to focus on 2 or 3 of the most appropriate and strongest areas. 

Inductive reasoning using Hill's Criteria is only one possible framework available to structure a Discussion. Other types of evidence and arguments could also contribute to putting the results of a study and the General Hypotheses that the results support into a broader scientific context.