Statistical Interpretation in Pilot Studies with Non-Significant Results
When p > 0.05 in a pilot study (n=50) examining the association between composite index of anthropometric failure and behavioral feeding problems, interpret this as preliminary evidence requiring a larger, adequately powered study rather than concluding no association exists.
Understanding the Statistical Context
Your pilot study's non-significant result does not mean there is no association between anthropometric failure and feeding problems—it likely reflects insufficient statistical power 1. With only 50 participants, your study may lack the sample size needed to detect a true association, even if one exists clinically 1.
Key Considerations for Interpretation
Effect size matters more than p-values in pilot studies: Report the actual effect size (odds ratio, correlation coefficient, or mean difference) with 95% confidence intervals, as these provide more meaningful information about the potential relationship than the p-value alone 1.
Clinical significance versus statistical significance: Research demonstrates that behavioral feeding problems are strongly associated with malnutrition, with total behavioral frequency scores ≥85 increasing malnutrition risk 3.7-fold (p<0.001) in adequately powered studies 2. Your non-significant result may reflect sample size limitations rather than absence of association 1.
Avoid dichotomous "positive/negative" classification: The practice of classifying studies as "positive" (p<0.05) or "negative" (p>0.05) is a misuse of p-values 1. Instead, use graded interpretation where smaller p-values indicate stronger evidence, but p>0.05 does not prove absence of effect 1.
Recommended Actions for Your Study
Report All Findings Transparently
Present descriptive statistics comprehensively: Report the prevalence of anthropometric failure using CIAF categories (wasting only, wasting and underweight, stunting only, etc.) and the distribution of feeding problem scores in your sample 3, 4.
Calculate and report effect sizes: Provide correlation coefficients or odds ratios with 95% confidence intervals, even when p>0.05, as these inform future sample size calculations 1.
Document feeding problem patterns: Use validated tools like the Behavioral Pediatrics Feeding Assessment Scale (BPFAS) to quantify both total behavioral frequency and total behavioral problem scores, as these have established associations with malnutrition in larger studies 2, 5.
Sample Size Considerations for Future Research
Conduct power analysis for definitive study: Use your pilot data's effect size estimates to calculate the required sample size for a properly powered study 1. Studies showing significant associations between feeding problems and malnutrition typically include 100-300 participants 2, 5, 3.
Consider the clinical context: Research in similar populations shows that 42.1% of children experience anthropometric failure when assessed by CIAF 3, and more than one-third of children with feeding problems demonstrate malnutrition 5. These prevalence rates should inform your power calculations.
Clinical Implications Despite Non-Significance
Evidence Supporting the Association
Established relationship in literature: Multiple studies demonstrate that behavioral feeding problems significantly correlate with anthropometric measures, including negative correlations with body weight (r=-0.338, p=0.012), BMI (r=-0.322, p=0.017), and mid-upper arm circumference (r=-0.384, p=0.004) 5.
CIAF captures comprehensive malnutrition: The composite index identifies multiple forms of anthropometric failure (stunting, wasting, underweight, and combinations) that single indicators miss 3, 4. Your study may have detected patterns in specific CIAF subcategories even if overall significance wasn't reached.
Feeding assessment provides actionable data: Observational assessment of feeding behaviors provides superior data to self-report measures, though it requires more resources 6. Consider whether your assessment method captured the full spectrum of feeding problems.
Practical Recommendations
Report nominal p-values with appropriate caveats: Present your p-value alongside effect sizes, clearly noting that the study was exploratory and underpowered for definitive conclusions 1.
Avoid overinterpretation: Do not conclude "no association exists" based solely on p>0.05 in an underpowered pilot study 1. Instead, state that your findings are inconclusive and require confirmation in a larger sample.
Consider subgroup analyses cautiously: If you observed trends in specific CIAF categories (e.g., stunting only versus combined failures) or specific feeding problem domains (child-related versus parent-related behaviors), report these descriptively without making inferential claims 2.
Methodological Refinements for Future Studies
Standardize assessment tools: Use validated instruments like BPFAS, which distinguishes total behavioral frequency from total behavioral problem scores and has demonstrated reliability in detecting feeding-related malnutrition 2, 5.
Control for confounding variables: Mother's height, family income, and child's age are dominant factors associated with anthropometric failure and should be included in multivariable analyses 3.
Consider multiple testing strategies: If examining multiple CIAF categories or feeding problem subscales, implement appropriate multiplicity correction methods (Bonferroni, Holm, or hierarchical testing) to control type I error 1.