# Bias in Veterinary Research: Types, Sources, and Mitigation


## Key Takeaways

- Selection bias, arising from non-representative sampling (e.g., referral populations, convenience sampling for necropsy), distorts findings by systematically excluding or over-representing certain cases or individuals, leading to estimates that do not reflect the true target population.
- Information bias, including observer bias and confirmation bias, systematically errors in measurement (e.g., subjective lameness scoring, histopathologic grading) due to lack of blinding, leading to inflated or deflated outcome estimates based on prior expectations.
- Confounding variables such as breed, age, or management system can create spurious associations between exposure and outcome if not addressed through design strategies like randomization, restriction, or matching.
- Sex and gender bias, characterized by the overrepresentation of one sex in study populations, limits the generalizability of findings, as seen in neurodevelopmental outcomes where sex-specific effects are documented.
- Geographic and taxonomic biases, such as research concentration in high-income regions or disproportionate attention to charismatic species, result in data deficiencies for understudied populations and contexts, impeding broad applicability of findings.
- Adherence to reporting standards like ARRIVE 2.0 and EQUATOR guidelines is crucial for transparency, making bias detectable by ensuring adequate detail on randomization, blinding, and sample size estimation.

---

Veterinary research informs clinical decisions, public health policy, and animal welfare standards. Yet every study design carries structural vulnerabilities that can distort findings, and the published literature itself reflects systematic preferences in what gets studied, where, and in which species. This article catalogues the principal forms of bias that affect veterinary research, from study conception through data interpretation, and describes practical strategies for minimizing their influence. It serves veterinary researchers designing studies, clinicians appraising the literature, and graduate students preparing protocols or systematic reviews.

The scope is deliberately cross-species. Bias operates similarly across companion animal, production animal, wildlife, and laboratory research, although the specific manifestations differ. Statistical adjustment methods are not covered here, the focus is on design, conduct, and reporting choices that prevent bias from entering a study in the first place. Where the evidence base is contested or incomplete, that uncertainty is stated directly.

## At a Glance

| Parameter | What the Reader Needs to Know |
|---|---|
| Selection bias | Occurs when study participants or samples are not representative of the target population, common in convenience sampling and referral populations |
| Information bias | Arises from systematic errors in measuring exposure, outcome, or covariates, includes recall bias, observer bias, and misclassification |
| Confounding | A third variable associated with both exposure and outcome that distorts the exposure-outcome relationship unless addressed in design |
| Confirmation bias | Tendency to interpret data in ways that support prior expectations, reduced by blinding, as demonstrated in animal behavior studies |
| Species bias | Disproportionate research attention to charismatic or economically important species, leaving others data-deficient |
| Sex and gender bias | Overrepresentation of one sex in study populations, limiting generalizability of findings across sexes |
| Geographic bias | Concentration of research in high-income regions, limiting applicability to other ecological and production contexts |
| Reporting standards | ARRIVE 2.0 and EQUATOR guidelines specify minimum information for transparent, reproducible research |

## The Logic of Bias in Study Design

Bias is a systematic error, not random variation. Random error diminishes precision and can be quantified with confidence intervals. Bias distorts the true value of an estimate in a consistent direction and cannot be corrected by increasing sample size. Understanding this distinction is foundational: a large, biased study produces a precise estimate of the wrong quantity.

The taxonomy of bias used in this article follows the conventional division into selection bias, information bias, and confounding. Selection bias distorts who or what enters the study. Information bias distorts what is measured about those participants. Confounding distorts the apparent relationship between exposure and outcome through an extraneous variable. Each operates at a different stage of the research process and each requires a different mitigation strategy.

## Selection Bias

Selection bias arises when the probability of inclusion in a study is related to both the exposure and the outcome of interest. In veterinary clinical research, the most common source is the referral population. Specialty hospitals see cases that differ systematically from the general patient population in severity, chronicity, prior treatment, and owner resources. A study of a therapeutic intervention conducted exclusively at a referral center may demonstrate efficacy in severe, refractory cases that does not extrapolate to first-opinion practice, or the reverse.

Sampling frames for wildlife and production animal studies carry analogous problems. Convenience sampling of animals presented for necropsy overrepresents fatal and severe disease. Passive surveillance systems detect cases that are tested, and testing is influenced by clinical suspicion, owner willingness, and laboratory access. Each of these pathways introduces a systematic filter between the true disease burden and the observed data.

The broader literature demonstrates that selection operates at the level of entire research programs, also individual studies. A systematic review of carnivoran research in China found pronounced species bias, with bears and big cats receiving the greatest attention while most small- and medium-sized carnivorans were largely neglected, and the giant panda attracting more than half of all research resources in the group. This pattern matters because data deficiency impedes conservation planning, and research bias may direct limited resources toward less threatened species or low biodiversity regions. Researchers should ask whether the species, population, or clinical subgroup they propose to study has been selected for scientific convenience instead of for the question at hand.

## Information Bias

Information bias results from systematic errors in measuring study variables. Observer bias occurs when the person assessing an outcome knows the treatment allocation and allows expectations to influence measurement. This is particularly problematic when outcomes involve subjective judgment, such as lameness scoring, histopathologic grading, or behavioral assessment.

The magnitude of this effect is not trivial. A meta-analysis of nestmate recognition studies in ants found that only 29% of 79 studies were conducted blind, and that blind studies were significantly more likely to report aggression among nestmates than non-blind studies. The authors attribute this to confirmation bias, the tendency to interpret information in a way that confirms prior expectations. Aggressive interactions between ants include subtle behaviors such as mandible flaring and recoil that are hard to quantify, making these assays prone to expectation-driven measurement error. Veterinary assessments that rely on similar subjective scoring systems carry the same vulnerability.

Blinding is the primary defense. Single blinding conceals treatment allocation from the outcome assessor. Double blinding conceals it from both assessor and participant, although in veterinary medicine the participant is the owner or handler instead of the animal. Blinding is feasible for most outcome measures, including histopathology, diagnostic imaging interpretation, and behavioral scoring, provided the study is designed with this requirement from the outset.

## Confounding

Confounding occurs when a third variable is associated with the exposure and independently affects the outcome, creating an apparent association that is not causal. In veterinary research, age, breed, sex, body condition, and management system are frequent confounders. A study examining the association between diet and a particular disease may find an apparent relationship that is actually mediated by breed predisposition, because breed influences both dietary choices and disease risk.

Confounding is addressed in the design phase through randomization, restriction, or matching. Randomization distributes both known and unknown confounders across treatment groups, which is why randomized controlled trials sit at the top of the evidence hierarchy for therapeutic questions. Restriction limits the study to a single stratum of the confounder, such as enrolling only animals of one breed, which improves internal validity at the cost of generalizability. Matching selects comparison subjects with similar values of the confounder.

## Sex and Species Bias in the Evidence Base

Sex bias operates across biomedical research. A review of gender differences in cancer symptom research documented that a large portion of both animal and human research has been, and continues to be, done primarily with male subjects, leading to inappropriate and questionable generalizations of research findings from males to females. In veterinary research, the parallel problem appears in studies that enrol only castrated males or only intact females for convenience, or that fail to report sex composition at all. Sex-specific effects are documented in neurodevelopmental outcomes following prenatal immune activation, where innate immune stimulation produces different behavioral and neurochemical consequences in males and females. Studies that ignore sex as a biological variable risk producing findings that apply to only half the target population.

Geographic bias similarly constrains the evidence base. A systematic review of field research on anthropogenic noise found a pronounced geographical bias, with most studies conducted in North America or Europe and a notable focus on terrestrial environments. Fewer than one-fifth of terrestrial studies were located in rural areas likely to experience urbanization by 2030, meaning data on ecosystems most likely to be affected by future changes are not being gathered. A parallel review of dogs as sentinels for human infectious disease found that 35% of studies described data from the Latin America and Caribbean region, 25% from North America, and only six of 142 studies described disease in Canada. Researchers should consider whether their findings will generalize across production systems, climates, and management practices, and should report the geographic and ecological context of their study populations explicitly.

## Reporting Standards as Bias Mitigation

Many biases enter the literature not through study conduct but through incomplete reporting. The ARRIVE guidelines, published by the NC3Rs, specify the minimum information required for transparent and reproducible animal research publications, including details of randomization, blinding, sample size estimation, and exclusion criteria. The EQUATOR Network maintains a comprehensive library of reporting guidelines, including CONSORT for randomized trials, PRISMA for systematic reviews, STROBE for observational studies, and REFLECT for livestock and food safety trials. Adherence to these standards does not prevent bias, but it makes bias detectable. A reader cannot assess the risk of bias in a study that does not report whether allocation was concealed or whether outcome assessors were blinded.

## Applied Bias Mitigation in Study Execution

### Pre-Registration and Protocol Locking

Pre-registration fixes the research question, primary outcome, analysis plan, and stopping rules before data collection begins. The protocol document serves as the reference against which deviations are identified and justified. For veterinary studies, the protocol should specify inclusion and exclusion criteria for animals, the method of allocation to treatment groups, the timing and technique of sample collection, and the statistical model planned for each outcome.

Protocol locking prevents two related biases. First, it blocks post hoc revision of hypotheses to match observed results, a practice that inflates false discovery rates. Second, it forces explicit decisions about handling missing data, outliers, and non-adherence before those problems appear in the dataset. When deviations become necessary, they should be documented with the date, the reason, and the expected effect on inference. The [ARRIVE 2.0 reporting guidelines](https://arriveguidelines.org/) require that publications state whether a study protocol was registered and where it can be accessed, which allows readers to assess the risk of selective reporting.

Species and production system change the practical details of pre-registration. Clinical trials in companion animals often recruit from a single hospital population, so the protocol must define how consecutive cases are screened and why some owners decline participation. Production animal studies may depend on seasonal disease outbreaks, requiring the protocol to specify trigger conditions for study initiation. Wildlife studies face unpredictable access to subjects, so the protocol should distinguish between planned analyzes and opportunistic sampling before data collection starts.

### Blinding and Masking Procedures

Blinding prevents expectation from influencing measurement, treatment administration, or outcome assessment. The [confirmation bias meta-analysis in nestmate recognition studies](https://pubmed.ncbi.nlm.nih.gov/23372659/) demonstrated the magnitude of this effect: blinded studies reported aggression among nestmates at 73% compared with 21% in unblinded studies. Veterinary researchers should assume that similar expectation effects operate in clinical scoring, histopathology grading, and behavioral assessment.

The minimum standard is blinding of outcome assessors. A technician who does not know treatment allocation should perform physical examinations, score lameness, read radiographs, or count lesions. When the same person must administer treatment and assess outcomes, a second observer should collect outcome data, or video records should be scored later by a masked reviewer. Blinding of animal handlers is often impossible in surgical or dietary studies, but the person measuring the primary outcome can almost always be masked.

Practical constraints differ by setting. In farm trials, ear tags or paint marks can encode treatment without revealing allocation to the assessor. In laboratory studies, cage cards can be coded so that husbandry staff remain unaware of group assignment. In wildlife field studies, blinding may be impossible for visible treatments such as ear notches or collars, but laboratory analyzes of collected samples can be blinded by using coded sample identifiers. The [systematic review of dogs as sentinels for human infectious disease](https://pubmed.ncbi.nlm.nih.gov/30248931/) noted that sample processing and diagnostic testing are stages where blinding is feasible even when field observations cannot be masked.

### Standardized Operating Procedures for Measurement

Measurement bias arises when data collection methods differ between groups or change over time. Standardized operating procedures (SOPs) reduce this risk by fixing the equipment, the anatomical site, the timing relative to treatment, and the recording format for each variable.

For clinical measurements, the SOP should specify calibration schedules for scales, sphygmomanometers, and ultrasound equipment. For laboratory assays, the SOP should name the assay kit, the sample type, the storage temperature, and the maximum time from collection to processing. For behavioral outcomes, the SOP should define the observation window, the ethogram, and the inter-observer reliability threshold. The [MSD Veterinary Manual](https://www.msdvetmanual.com/) provides species-specific reference ranges and examination techniques that can anchor these definitions, though researchers must confirm that the chosen method matches the reference population.

Equipment availability changes the correct choice of measurement strategy. A study in a referral hospital may use CT for tumor staging, while a practice-based study may rely on radiography and cytology. The protocol should specify the minimum diagnostic standard for inclusion and record which modality was used for each case, so that subgroup analyzes can account for measurement differences. Similarly, point-of-care analyzers may be the only option in field settings, and the SOP must document the device, the quality control schedule, and the correlation with reference laboratory methods if both are used.

### Documentation and Audit Trails

Complete documentation supports bias detection during analysis and peer review. Each animal should have a study record that includes enrollment date, allocation, all measurements with timestamps, protocol deviations, and adverse events. The record should distinguish between raw data and derived variables, and the derivation formulas should be written in the analysis code or a data dictionary.

Audit trails become critical when data are entered manually. Double data entry for a random sample of records quantifies entry error rates. Range checks and logic checks in the database flag implausible values, such as a body weight recorded as 450 kg for a cat or a temperature of 41°C in a healthy control. The [EQUATOR Network reporting guidelines](https://www.equator-network.org/) include items on data management and quality control that apply across study designs, and reviewers increasingly expect these details in published manuscripts.

Documentation standards vary by sector. Clinical trials in regulated environments may require source data verification against medical records. Production animal studies should retain batch records for feed, vaccines, and medications. Wildlife studies should archive field notebooks, GPS tracks, and photographic evidence. The [WOAH terrestrial animal health standards](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) specify documentation requirements for surveillance and diagnostic studies that support international trade, and these can serve as a model for data quality in other veterinary research contexts.

### Monitoring Parameters and Interim Checks

Interim monitoring detects bias as it emerges instead of after study completion. The monitoring plan should specify the frequency of data review, the responsible person, and the criteria for stopping or modifying the study.

| Monitoring parameter | What it detects | Action threshold | Consequence |
| --- | --- | --- | --- |
| Recruitment rate per site | Selective enrollment or referral bias | Rate falls below 50% of projected | Investigate site procedures, retrain recruiters |
| Baseline balance on key covariates | Randomisation failure or allocation bias | Imbalance > 20% on any prognostic variable | Stratified analysis, consider covariate adjustment |
| Missing data proportion | Differential follow-up or measurement failure | > 10% of primary outcome data missing | Compare missingness by group, plan sensitivity analysis |
| Inter-observer agreement on subjective scores | Drift in measurement standards | Kappa < 0.60 | Retrain assessors, re-score archived material |
| Protocol deviation rate | Execution bias | > 5% of enrolled animals | Document deviations, analyze by intention to treat |
| Adverse event frequency | Safety signal or differential reporting | Event rate differs between groups | Unblind safety committee, consider early stopping |

The thresholds in this table are starting points, not universal rules. A multicentre trial may tolerate a higher deviation rate than a single-center study because the case mix is more variable. A wildlife study with unpredictable capture success may need a lower recruitment threshold because the alternative is no study at all. The monitoring plan should be written before data collection begins, and the thresholds should be justified in the protocol.

Interim checks are distinct from interim analyzes of efficacy. Efficacy analyzes require formal stopping rules and statistical correction for repeated looks at the data. Monitoring checks focus on data quality and operational performance, and they do not require adjustment of significance levels unless they trigger a formal interim analysis. The distinction should be stated in the protocol to prevent confusion during study conduct.

The [AVMA practice resources](https://www.avma.org/resources-tools) include guidance on clinical research conduct that addresses documentation, adverse event reporting, and data integrity. These resources are oriented to United States practice, so researchers in other jurisdictions should confirm which standards apply locally. The core principle is universal: the study record must be complete enough that an independent reviewer can reconstruct every decision that affected the data.

## Recognized Failure Modes and Early Detection

Bias mitigation strategies fail in characteriztic patterns. The most common is protocol drift, where measurement procedures degrade gradually across the study period. Detect this by tracking operator-specific means and variances in calibration checks. If one technician's readings shift mid-study while others remain stable, retraining and data review are indicated before further enrollment.

A second failure mode is unmasking, whether accidental or through subtle cues. Animals that receive an intervention may behave differently, smell differently, or carry injection-site reactions that reveal allocation. Early detection requires structured checks: ask blinded assessors at intervals whether they believe they know the allocation, and record their confidence. If correct guesses exceed chance substantially, masking has failed and the analysis plan must account for this.

Attrition bias emerges when withdrawal rates differ between groups. Monitor dropout counts and reasons at each interim review. Differential attrition that correlates with treatment assignment threatens validity more than overall attrition does. The discriminating check is a comparison of baseline characteriztics between completers and non-completers within each arm.

## Common Errors and Corrective Action

Less experienced investigators frequently conflate randomisation with allocation concealment. Randomisation without concealment still permits selection bias, because enrolling staff can influence which animal receives which intervention. The corrective action is to separate the person generating the allocation sequence from the person enrolling animals, and to use opaque, sequentially numbered containers or a central randomisation service.

A second recurring error is the inclusion of multiple observations from the same animal without accounting for clustering. Repeated measurements on one subject are not independent data points. Students and early-career researchers often treat each measurement as an independent observation, which inflates sample size and narrows confidence intervals. The corrective action is to specify the experimental unit at the design stage and to analyze at that level, or to use methods that accommodate within-subject correlation.

A third error is over-interpreting subgroup analyzes. When a primary analysis shows no effect, investigators may search subgroups for a positive finding. This practice invites type I error. The corrective action is to pre-specify subgroups and their direction of effect in the protocol, and to treat post hoc subgroup findings as hypothesis-generating only.

## Limitations of the Current Evidence

The veterinary evidence base carries structural limitations that bias mitigation cannot fully correct. Publication bias favours positive results, and this operates at the level of journals, funders, and investigators. Systematic reviews that rely on published literature inherit this distortion. The [EQUATOR Network reporting guidelines](https://www.equator-network.org/) provide a library of standards that improve transparency, but they cannot recover studies that were never submitted or never accepted for publication.

Geographic and taxonomic bias further constrains generalizability. Research on anthropogenic noise effects concentrates in North America and Europe, with few studies in regions projected to experience rapid urbanisation, and taxonomic coverage is uneven [field research on anthropogenic noise shows pronounced geographic and taxonomic bias](https://pubmed.ncbi.nlm.nih.gov/32277776/). Similarly, research effort on carnivorans in China is heavily skewed toward charismatic species such as bears and big cats, leaving small- and medium-sized carnivorans understudied [species bias in carnivoran research in China](https://pubmed.ncbi.nlm.nih.gov/33998183/). Findings from one species or region may not transfer to another.

Sex bias persists across biomedical research. Studies that enrol only males and generalize to females risk missing sex-specific effects, particularly in neurodevelopmental and immune outcomes [sex effects on neurodevelopmental outcomes of innate immune activation](https://pubmed.ncbi.nlm.nih.gov/22516179/). The same concern applies to veterinary species, where sex, neuter status, and breed can modify treatment responses.

Expert opinion still differs on several points. Whether blinding is feasible or necessary for all outcome types, how strictly to apply intention-to-treat principles in veterinary field studies, and how to weight evidence from observational studies against randomised trials all remain contested. The [ARRIVE guidelines](https://arriveguidelines.org/) specify minimum reporting requirements, but they do not resolve these design debates.

## Escalation and External Consultation

Certain circumstances warrant escalation beyond the study team. If interim monitoring reveals systematic measurement error that cannot be corrected by retraining, consult a laboratory or measurement specialist before continuing. If differential attrition threatens the validity of a trial, seek statistical consultation to determine whether the analysis plan can accommodate the loss or whether the study must be stopped.

Regulatory reporting obligations arise when research identifies notifiable disease, unexpected adverse events, or welfare concerns that exceed approved thresholds. The [WOAH terrestrial animal health standards](https://www.woah.org/en/what-we-do/standards/codes-and-manuals/terrestrial-code-online-access/) define international notification requirements for listed diseases, and national authorities may impose additional obligations. Investigators should confirm applicable requirements before commencing work, because reporting timelines are typically short.

Ethical review boards and institutional animal care committees should be consulted whenever the protocol changes in ways that affect animal welfare, including changes to endpoints, sample sizes, or procedures. The [AVMA professional practice resources](https://www.avma.org/resources-tools) provide guidance on welfare standards and professional obligations that complement institutional oversight.

## Troubleshooting Table

| Observation | Likely Cause | Discriminating Check |
| --- | --- | --- |
| Blinded assessor guesses allocation correctly >70% of the time | Unmasking through behavioral or physical cues | Interview assessors about which cues they used, compare guess accuracy against chance |
| Technician measurements drift mid-study | Protocol drift or equipment calibration failure | Plot measurement means by technician and date, rerun calibration standards |
| Dropout rates differ markedly between groups | Differential attrition related to treatment burden or adverse effects | Compare reasons for withdrawal across arms, examine baseline characteriztics of dropouts |
| Subgroup finding appears only after primary analysis | Post hoc exploration inflating type I error | Check whether subgroup and direction were pre-specified in the protocol |
| Repeated measures treated as independent observations | Misidentification of the experimental unit | Review the analysis plan, confirm whether clustering was accounted for |
| Positive results dominate the available literature on a topic | Publication bias | Search trial registries and grey literature for unpublished or null studies |

## Frequently Asked Questions

### How do I manage bias mitigation when funding or time is limited?

Prioritize blinding and random allocation before investing in expensive equipment. These two measures address the most pervasive biases and cost little beyond planning effort. Use simple randomisation with sealed envelopes or a web-based sequence generator. For measurement, standardize the protocol on one page and train all assessors together. If full blinding is impossible, blind the outcome assessment at minimum, since that step prevents confirmation bias from shaping data interpretation. The ARRIVE guidelines list blinding as essential information for transparent reporting, so reviewers will expect it regardless of budget. Document every compromise in the methods section and discuss the residual risk in the limitations paragraph.

### What should I do when true blinding is impossible in my study design?

Use objective outcome measures that leave little room for observer interpretation. Automated readouts, quantitative laboratory assays, and validated scoring systems reduce the influence of expectation when the investigator must know the treatment group. For behavioral or histopathological scoring, have a second observer who is blinded to the hypothesis score a random subset of samples, then calculate inter-observer agreement. If the study involves client-owned animals, consider separating the clinician delivering treatment from the clinician assessing outcomes. When even that separation fails, state the limitation explicitly and cite the evidence that non-blinded studies over-report expected effects, as demonstrated in nestmate recognition research where blind studies more often detected aggression among nestmates.

### How does bias mitigation differ between companion animal and production animal research?

Companion animal studies face owner expectation bias and selective enrollment, since owners may decline randomisation to a perceived inferior treatment. Production animal research operates at herd level, where unit of allocation errors and confounding by management practices dominate. In companion animal work, train owners to use standardized diaries and consider placebo controls where ethically permissible. In production settings, allocate at the pen or herd level, match groups on baseline production parameters, and record management events that occur during the study period. Species-specific reporting guidelines exist within the EQUATOR Network library, and the WOAH terrestrial animal health standards provide frameworks for surveillance studies that cross production systems.

### What records should I keep to support a bias audit after study completion?

Keep the pre-registration document with date stamps, the randomisation sequence and allocation log, all versions of the protocol with revision dates, training records for personnel who collected data, and the raw data files with metadata describing collection conditions. Maintain a deviation log that records every instance where the protocol was not followed, including the date, the person responsible, and the corrective action taken. Retain communication records that document decisions about exclusions or data handling. These materials allow a reviewer or external consultant to reconstruct the study flow and verify that masking was preserved. The ARRIVE guidelines specify which items should appear in the final publication, but the underlying audit trail should be kept for at least the data retention period required by your institution.

### How do I explain study limitations to a referring veterinarian or practice owner?

Describe the limitation in terms of clinical consequence instead of statistical abstraction. State what the study can and cannot support, and name the specific bias that threatens the conclusion. For example, explain that the study was not blinded, so the assessor's expectation may have influenced subjective scores, and therefore the reported effect size is uncertain. Offer the practical implication: the treatment may work, but the magnitude of benefit is less certain than the paper suggests. Direct them to the MSD Veterinary Manual for balanced summaries of therapeutic evidence and to the AVMA practice resources for guidance on interpreting research in clinical decisions. Avoid overstating certainty, and acknowledge when the evidence base is genuinely contested.

### When should I consult a statistician or epidemiologist about bias instead of proceed alone?

Consult before finalising the protocol whenever the study involves multiple sites, repeated measurements over time, or outcomes that require subjective judgment. Seek external input when you cannot articulate how each potential confounder will be handled, or when the allocation unit differs from the analysis unit. A statistician can also help when you plan to combine data from different sources or when the sample size calculation depends on assumptions you cannot verify. Early consultation costs less than correcting a flawed design after data collection. The EQUATOR Network reporting guidelines can help you identify which methodological details reviewers will scrutinise, and discussing those items with a colleague before starting the study often reveals gaps in the planned bias mitigation.

## Related Clinical & Scientific Guides

* [Conducting Systematic Reviews of Veterinary Diagnostic Test Accuracy](/knowledge/veterinary-medicine/veterinary-research-methods/conducting-systematic-reviews-veterinary-diagnostic-test-accuracy)
* [Cluster Randomized Trials in Veterinary Research: Design and Analysis](/knowledge/veterinary-medicine/veterinary-research-methods/cluster-randomized-trials-veterinary-research-design-analysis)
* [Performing Economic Evaluations of Veterinary Interventions](/knowledge/veterinary-medicine/veterinary-research-methods/performing-economic-evaluations-veterinary-interventions)


## References and Further Reading

- [Confirmation bias in studies of nestmate recognition: a cautionary note for research into the behavior of animals.](https://pubmed.ncbi.nlm.nih.gov/23372659/). 2013.
- [Species bias and spillover effects in scientific research on Carnivora in China.](https://pubmed.ncbi.nlm.nih.gov/33998183/). 2021.
- [Gender differences in pain, fatigue, and depression in patients with cancer.](https://pubmed.ncbi.nlm.nih.gov/15263057/). 2004.
- [Dogs (<i>Canis familiaris</i>) as Sentinels for Human Infectious Disease and Application to Canadian Populations: A Systematic Review.](https://pubmed.ncbi.nlm.nih.gov/30248931/). 2018.
- [Sex effects on neurodevelopmental outcomes of innate immune activation during prenatal and neonatal life.](https://pubmed.ncbi.nlm.nih.gov/22516179/). 2012.
- [Trends and knowledge gaps in field research investigating effects of anthropogenic noise.](https://pubmed.ncbi.nlm.nih.gov/32277776/). 2021.
- [ARRIVE Guidelines 2.0 for Reporting Animal Research](https://arriveguidelines.org/). PLOS Biology, 2020.
- [EQUATOR Network Reporting Guidelines](https://www.equator-network.org/). EQUATOR Network.
- [MSD Veterinary Manual, Professional Edition](https://www.msdvetmanual.com/). MSD Veterinary Manual.

## Related Articles

- [Using Mixed Methods in Veterinary Research](/knowledge/veterinary-medicine/veterinary-research-methods/using-mixed-methods-veterinary-research)
- [Conducting Qualitative Research in Veterinary Settings](/knowledge/veterinary-medicine/veterinary-research-methods/conducting-qualitative-research-veterinary-settings)
- [How to Write a Research Protocol for Veterinary Studies](/knowledge/veterinary-medicine/veterinary-research-methods/write-research-protocol-veterinary-studies)
- [Applying Competing Risks Analysis in Veterinary Research](/knowledge/veterinary-medicine/veterinary-research-methods/applying-competing-risks-analysis-veterinary-research)
- [The ARRIVE Guidelines for Animal Research: Implementation and Compliance](/knowledge/veterinary-medicine/veterinary-research-methods/arrive-guidelines-animal-research-implementation-compliance)

> This article is educational professional reference material for veterinary audiences. It is not a substitute for veterinary diagnosis, individual clinical judgment, current product labeling, or applicable regulatory requirements.