Drawing Evidence-Based Conclusions from Reproduction and Inheritance Data

Hello! Last lesson, you used pedigrees and offspring ratios to infer inheritance patterns. This lesson asks a related but broader question: what can experimental results actually justify? You’ll analyse counts from controlled reproduction and inheritance experiments, compare observations with predictions, and write conclusions that are strong without claiming more than the evidence shows. Allow about 40 minutes.

Start with what the experiment measured

Mendel’s pea experiments were informative not simply because he counted offspring, but because he controlled which plants reproduced and recorded the results over generations. Watch this short portion of Mendelian Genetics and Punnett Squares by Professor Dave Explains with the experimental setup in mind.

Before interpreting any dataset, identify its comparison and its measured outcome. Consider a hypothetical experiment comparing two ways of pollinating the same variety of flowering plant. In each of three trials, a researcher pollinates 30 flowers with pollen from another plant and 30 with pollen from the same plant. The researcher records how many flowers develop seed-bearing pods.

The outcome is flowers that developed seed-bearing pods, not the number of seeds and not the number of offspring that survived. Across the three trials, the success rates were $72/90=80\%$ and $48/90\approx 53\%$, a difference of about 27 percentage points. All three trials show higher pod development with pollen from another plant. Those are useful pieces of evidence: the size of the difference and its repetition across trials.

A defensible conclusion is that, under these experimental conditions, pollination with pollen from another plant was associated with a higher proportion of flowers developing seed-bearing pods. It would be too strong to conclude that self-pollination never works: 48 self-pollinated flowers did develop pods. Nor do these counts alone reveal why the difference occurred. If the two treatments received different amounts of pollen or were handled differently, those differences could also affect pod development. Keeping flower conditions, pollen delivery and handling comparable makes the reproductive comparison more convincing.

Notice that this is not a Mendelian offspring-ratio question. There is no reason to test these two pod-development percentages against a 3:1 ratio. The appropriate analysis follows from what the experiment measured.

Compare inheritance counts with a prediction

For an inheritance experiment, a proposed parental cross can provide an expected ratio before the offspring are counted. Real counts rarely match that ratio exactly: each offspring is a chance outcome. The question is whether the mismatch is reasonably consistent with chance variation under the proposed model.

Read The chi-squared Test from Biology LibreTexts for a worked example. Its tail-less mouse data provide a useful contrast with the modest variation discussed last lesson.

Here is the logic in the mouse example. Tail-less mice are $Tt$; the proposed cross is $Tt\times Tt$, with $T$ producing the tail-less phenotype under the initial simple-dominance model. The predicted offspring phenotypes are 3 tail-less:1 normal. Of 229 observed offspring, 139 were tail-less and 90 were normal.

An expected count is the total multiplied by the predicted fraction. It is not a count taken from the results.

Adding the final column gives a chi-squared statistic of approximately $24.98$. Squaring prevents a shortfall in one category from cancelling an excess in the other; dividing by $E$ accounts for the size of each expected category.

The null hypothesis here is that the proposed 3:1 model is correct and departures from it arise from random sampling. With two phenotype categories, the degrees of freedom are $2-1=1$. At the commonly used $0.05$ significance threshold, the critical chi-squared value for one degree of freedom is approximately $3.84$. Because $24.98$ is greater than $3.84$, the observed counts differ significantly from the 3:1 prediction. Reject that particular null hypothesis.

That decision needs careful wording. A $p$-value is the probability, assuming the null model, of obtaining a result at least as far from its prediction as the one observed. It is not the probability that the null model is true, nor does a statistically significant result identify the cause of the mismatch. Conversely, failing to reject a model would not prove it correct; several models might fit the same counts.

Turn a mismatch into a bounded conclusion

The mouse results are not close to the predicted counts: the model expects about 172 tail-less and 57 normal offspring, but the researchers counted 139 and 90. The evidence tells you that simple dominance with all predicted offspring represented equally among the counted offspring is a poor fit. It does not, on its own, tell you what happened to any missing offspring.

The follow-up described in the reading found that $TT$ embryos die. This changes which genotypes can appear among the offspring being counted. A $Tt\times Tt$ cross initially predicts $1TT:2Tt:1tt$. If $TT$ embryos do not survive, the predicted ratio among surviving offspring becomes $2$ tail-less $Tt$ to $1$ normal $tt$. Thus, when researchers count matters: a ratio among surviving young need not represent the ratio at fertilisation.

The distinction between evidence and explanation is central. The 139:90 counts prompted investigation because they conflicted with 3:1. Evidence about embryo survival was needed to support the explanation involving $TT$. An alternative model suggested after looking at the counts is worth investigating, but the original mismatch alone does not prove it.

For an HSC response, you can make that reasoning compact without leaving out the qualification:

The offspring data do not support the predicted 3:1 ratio for a simple $Tt\times Tt$ cross among counted offspring. Of 229 mice, approximately 172 tail-less and 57 normal mice were expected, but 139 and 90 were observed. The chi-squared value of $24.98$ exceeds the $0.05$ critical value of $3.84$, so random sampling under that model is an unlikely explanation for a mismatch this large. Further evidence is needed to determine its cause.

This structure works even when no statistical test is requested: state the conclusion, select the relevant comparison, and limit the claim to what was measured. For the pollination experiment, that means comparing proportions of flowers developing pods. For the mouse experiment, it means comparing phenotype counts with a prediction and distinguishing inheritance from survival.

Key takeaways

Experimental data become meaningful when you know what was counted, what was compared, and what the proposed model predicted. Use totals and proportions for comparisons between treatments; use expected counts when testing an inheritance ratio. A mismatch can justify rejecting a specified prediction, but it does not automatically establish a particular biological explanation.

This completes the heredity module. Next, you’ll move from how alleles are transmitted to how genetic material can change, beginning with gene, chromosome and genome mutations.