A controversy in evolutionary genetics is whether active dosage compensation is necessary to resolve the gene dosage imbalance between the X chromosome and autosomes. ScRNA-seq data could provide insight into this issue. However, it's crucial to carefully evaluate whether inherent characteristics of scRNA-seq, such as the sparsity of detected genes, might bias the X:AA expression ratio in mammals. This study evaluated two common strategies for selecting genes in the calculation of X:AA, namely, filter-by-expression and filter-by-fraction, with simulated scRNA-seq and bulk RNA-seq datasets. We found that both strategies produce an inflated X:AA, thus artifactually supporting dosage compensation. Analyzing empirical human Smart-seq2 data, results from the filter-by-expression strategy suggested that X-linked genes were more highly expressed than autosomal genes, a pattern that is neither predicted by dosage compensation nor explained by genes escaping X chromosome inactivation. However, the results of the filter-by-fraction strategy are consistent with the simulation. Furthermore, despite biasing for mean expression levels, we found that scRNA-seq data could be used to detect X-to-autosome expression noise differences as small as 10%, which enabled investigation into the distribution of genes that are more likely insensitive to gene dosage changes. Analysis of the empirical Smart-seq2 data revealed a 10% to 15% increase in expression noise for X chromosomes compared with autosomes and a significant depletion of dosage-sensitive genes on X chromosomes. Overall, these results highlight the need to be cautious when interpreting scRNA-seq data, particularly when comparing the expression of different genes, and provide additional evidence for the hypothesis of X chromosome insensitivity.