# Density plot looks normal, qqplot not normal Shapiro significative

The following QQ plot looks with too many points out of the line, the density plot looks normal and the Shapiro Test p-value < 2.2e-16, so this is not a normal distribution but I've read not to trust Shapiro Test when I have about 1000 data points so I should conclude that this distribution is normal?

• Clear asymmetry suggests something is up. A) Did you calculate the skewness? A nominal rule-of-thumb is if Pearson Skewness $\ge$ 0.1, then you have to take corrective action, e.g., performing statistics on the log() of your measurements, rather thanon the measurements direclty. Also, are there additional factors (meta data, etc.) that you can use to subselect your data? This may also be a mixture of models, so you might have two or 3 normal distributions all sitting close together, but the second and third are small enough so as not to create an obviously multimodal histogram. – Peter Leopold Feb 17 at 16:07
• @Peter Where does that rule of thumb come from? It's not generally applicable, so it would be of interest to know its limitations and assumptions. – whuber Feb 17 at 16:12
• @PeterLeopold "Pearson skewness" is not uniquely defined. Pearson himself put most emphasis on measuring skewness relative to the mode, which had a major role in his system of distributions. But he did also use a dimensionless ratio based on third and second moments around the mean. And yet again (mean $-$ median) / SD appears in his work. But regardless I wouldn't regard skewness of about 0.1 on any measure I've encountered as requiring transformation. I would always want to see the data, however. – Nick Cox Feb 17 at 16:18
• @Peter The problem here is more about kurtosis than skewness and it's not clear that taking the log is justified, even if it were skew. It depends on what the OP is going to use the data for. – Peter Flom Feb 17 at 16:18
• I would not trust a scale for happiness [NB] with such results! – Nick Cox Feb 17 at 16:20