C2C Original Research

Did the Virus or Governments Kill More Canadians? A Statistical Analysis of Excess Deaths during the Pandemic

Jim Mason
January 31, 2023
After talking about little but Covid-19 for two-and-a-half years, governments, academia and media became strangely incurious about the net effects of the pandemic on the overall mortality of Canadians. It seems a fundamental question worth asking. Learning the true toll taken by the disease and the measures meant to contain it is vital to assessing the efficacy and rationality of our responses – and whether we should do things differently next time. Using raw data from the federal government, Jim Mason, PhD, performed extensive statistical analysis of Canada’s death dynamics before and during the pandemic. Did Covid-19 raise Canada’s death numbers above normal? Was the government-imposed cure worse than the disease?
C2C Original Research

Did the Virus or Governments Kill More Canadians? A Statistical Analysis of Excess Deaths during the Pandemic

Jim Mason
January 31, 2023
After talking about little but Covid-19 for two-and-a-half years, governments, academia and media became strangely incurious about the net effects of the pandemic on the overall mortality of Canadians. It seems a fundamental question worth asking. Learning the true toll taken by the disease and the measures meant to contain it is vital to assessing the efficacy and rationality of our responses – and whether we should do things differently next time. Using raw data from the federal government, Jim Mason, PhD, performed extensive statistical analysis of Canada’s death dynamics before and during the pandemic. Did Covid-19 raise Canada’s death numbers above normal? Was the government-imposed cure worse than the disease?
Share on Facebook
Share on Twitter

Two metrics figured most prominently throughout the Covid-19 pandemic: the number of reported cases and the number of deaths attributed to the virus. That thousands of Canadians died directly of Covid-19 is indisputable, and governments have published official statistics for deaths from this cause. It is also certain that people died from other causes brought about by the pandemic or governments’ response to it, such as deferred treatment for non-Covid-19 ailments or drug overdoses from despair. Estimating that total with confidence would be very difficult, especially since the primary cause of many such deaths would remain debatable or inconclusive. Lastly, it is reasonable to assume that the pandemic or responses to it averted some deaths – such as from the drop in motor vehicle use in spring/summer 2020. This figure, too, would be very difficult to estimate with confidence.

Where does this leave those interested in assessing the overall human toll taken by Covid-19, directly and indirectly, and the overall costs or benefits, as measured in human fatalities, of governments’ response to the pandemic? Fortunately, there is a measurement that, while challenging to calculate and ultimately imperfect, provides a close approximation: excess deaths. Excess deaths are those over and above the number that would be expected based on total deaths observed in previous years, absent any unusual circumstances.

SARS-CoV-2 claimed thousands of Canadian lives beginning in early 2020. But did the virus actually result in “excess deaths” – more annual mortality than would be predicted from previous death statistics? And what about the effects of government pandemic measures? Did lockdowns and enforced distancing avert a large number of deaths? Or did the negative consequences cause additional deaths? The author sought evidence through careful statistical analysis. (Source of bottom left photo: Shutterstock)

Previous C2C Journal research articles have examined the implications of Covid-19 case rates as a function of vaccination status (as published on the Government of Ontario’s Covid-19 webpages), in particular the collapsing effectiveness of the vaccines that these data demonstrate and the lack of evidentiary support these data provide for coerced vaccination and vaccine passports.

This research essay examines whether the pandemic caused excess deaths in Canada and, if so, whether these deaths can be unequivocally attributed to the virus or whether some might have been the consequence of governments’ responses to the virus. This analysis uses Canada-wide data published on the Statistics Canada website.

The Foundation – Estimating What is Expected

Foundational in this kind of analysis is estimating the study area’s expected number of deaths had there been no abnormal event. Without a reliable estimate of this number, any conclusions about excess deaths are purely speculative.

The data used to estimate the expected number of deaths will have some intrinsic variability that introduces some uncertainty into the estimated expected number. This complicates the task of distinguishing apparent from actual excess deaths. Assessing the significance of an apparent number of excess deaths requires a measurement for the aforementioned uncertainty. If the uncertainty in the predicted number of deaths is large, then a seemingly large number of excess deaths may not, in fact, be significant.

This is why the first few figures below concern estimating the number of deaths that would have been expected in Canada in the first two pandemic years based on observed deaths during preceding years and calculating the range of uncertainty associated with these estimates that result from the historic data. Only then can we begin to evaluate whether, when and to what degree Canada experienced excess deaths.

Methodology and Statistical Approach

This original analysis used Statistics Canada data on Canada-wide monthly numbers of deaths from all causes. The analysis was conducted in four ways: on a monthly basis looking at each month individually; on a monthly basis looking at the cumulative number of deaths to the end of each month; on an annual basis looking at the total number of deaths in each calendar year; and on a seasonal basis using the numbers of deaths during the October-May “flu season” that were over and above the background trend from summer to summer.

Expected numbers of deaths were determined by extrapolating to 2020 and 2021 a best-fit straight line using the observed values from the preceding decade. In some cases, where the data clearly indicated linear behaviour from 2013-2019 but non-linear behaviour before 2013, the period 2013-2019 was used for determining the straight line. Otherwise the full 2010-2019 dataset was used. Using the goodness-of-fit measure (R2) for the straight line, uncertainty ranges (95 percent confidence intervals) were determined for each predicted data point in order to assess the significance of any deviation of the actual value from this prediction.

The analysis presented here does not assume any fundamental probabilistic model. It predicts expected numbers of deaths during the pandemic years through straightforward extrapolation of actual numbers observed and trends clearly present in the pre-pandemic data.

The raw data for this analysis are StatsCan’s annual and monthly actual death counts from 2010 through 2020 plus, for 2021, StatsCan’s provisional “adjusted” death counts (counts equivalent to the 2010-2020 dataset were not yet published when this analysis was conducted; this may have changed by the publication date of this article).

To achieve the smallest uncertainty in the predicted number, the analysis used as much of the data from 2010 to 2020 that could be reliably used without potentially distorting the prediction. Where less than the entire amount is used, the rationale for doing so is discussed in the text. This approach allows the nature of the data to drive the analysis rather than forcing the data to fit a presupposed theoretical model.

Around the world, numerous analyses of excess deaths during the pandemic have been conducted. Other than this analysis, to date only three have covered Canada: one by StatsCan, one by Our World in Data (OWID), both of which are discussed in Appendix A, and one in the Canadian Medical Association Journal, discussed lower down.

Although no a priori assumptions were made about the trends in the data, as will be discussed, the data seem to be well-described using linear relationships. This approach is the same as OWID’s except that this analysis used a longer history to derive the projected number of deaths. Finally, the StatsCan and OWID analyses were done purely on a weekly basis, in contrast to the four separate temporal perspectives prepared and evaluated in this analysis.

Seasonal Perspective

The process of evaluation begins with the seasonal analysis because this reveals some aspects of the data that affect the other analyses. Figure 1 charts the monthly death counts for Canada from January 2010 to December 2021. There are two obvious features:

  1. A general upward trend, consistent with a general upward trend in Canada’s population (also shown in Figure 1). More people result in more deaths.
  2. A repetitive and strikingly consistent pattern: June through September are roughly the same in every year, although increasing from year to year, while October through May are always higher than June through September, creating a similar but varying jagged-peak shape year after year.
 

The overall pattern resembles the cross-section of a gently upward-sloping plain with several mountain ranges sitting on top of it (Figure 2).

Figure 1. The foundation: Number of deaths by month from all causes in Canada from January 2010 through December 2021, showing a general increasing trend consistent with the trend of increasing population, and recurring 12-month oscillations. (Chart by Jim Mason)
Figure 2. The change in monthly deaths over time shows two underlying linear trends, one from January 2010 through October 2012, the other from October 2012 through December 2021. These define two gently sloping “plains” on which recurring, jagged “mountain ranges” are superimposed. The grey trend line is defined using data from October 2012 through December 2019, the pre-pandemic years, then used to predict values for 2020 and 2021, the pandemic years. (Chart by Jim Mason)

While the upward trend looks pretty constant, careful inspection shows that there are two different sections. The plains sections in 2010, 2011 and 2012 form a nearly horizontal line (green line in Figure 2). From 2013 to 2019 the plains sections are also straight but sloping slightly upward. The difference in the slopes is small but clear when extended across the entire time period (grey line in Figure 2). Although the grey line extends into the pandemic years, these years are excluded from the slope analysis because those years required calculation of predicted deaths.

The October through May “mountain ranges” include the normal flu season and also correspond to the periods during which large numbers of deaths were attributed to Covid-19. They have a remarkably consistent profile (Figure 3) except for the 2019-2020 and 2020-2021 seasons, which are shown in colour.

The 2019-2020 season has a higher number of deaths during April and May 2020 but a lower number of deaths during the preceding December, January and February. Since Covid-19 measures did not start until March 2020, this could indicate a milder-than-normal flu strain during this period.

The 2020-2021 season has higher numbers of deaths during October-December 2020 but lower numbers during January-April 2021. Since this is well into the pandemic, it is not possible from these data to distinguish flu deaths from Covid-19 deaths. One interpretation of this profile is that the seeming excess deaths in October-December 2020 are deaths that normally would have occurred during the following January-April but occurred earlier due to the complications resulting from the interaction of Covid-19 with comorbidities.

Figure 3. When plotted over the same September to June months, the annual “mountain range” death phenomenon shows a remarkably consistent shape throughout the last decade – except in the two pandemic years. Note: “Mountain range” deaths are the number of deaths exceeding the predicted monthly baseline depicted in Figure 2. (Chart by Jim Mason)

One possibility for analyzing excess deaths during the pandemic is by comparing the number of seasonal mountain range deaths in the 2019-2020 and 2020-2021 seasons (subtracting the underlying plains deaths) to the number expected by extrapolating the pre-pandemic mountain range deaths. The mountain range deaths charted in Figure 4 show an apparently linear upward trend over the study period. This is consistent with an increasing population and with an aging population.

Figure 5 charts changes in the fraction of Canada’s population aged 50+, the demographic that accounted for over 97 percent of the reported Covid-19 deaths (60 percent of these deaths were in the 80+ group.) Figure 5 also shows the impact this increasing fraction of older people had on the median age of the population, another way of showing that, overall, the population is aging.

Figure 4. The total number of deaths in each year’s September-June mountain range (blue), as graphed in Figure 3, shows a clear upward linear trend (green) over the last decade, with an almost regular year-to-year oscillation in the seasonal total. The pandemic year mountain range totals are both high, but still within the range of variation that would be expected, statistically, based on this history. The variation is indicated by the black bars, with a range of uncertainty that would include 95 percent of possible predicted values (or 95 percent confidence). (Chart by Jim Mason)
Figure 5. Canada’s population has been aging over the last decade, which should result in higher mortality, both as a numerical total and per million population. The proportion of the population aged 50+ has been increasing at a greater rate than the median age. This demographic accounted for 97 percent of deaths attributed to Covid-19. (Chart by Jim Mason)

The green line in Figure 4 represents the best-fit straight line to the data points in the pre-pandemic seasons of 2010-2011 through 2018-2019. This is used to predict the number of deaths expected in the 2019-2020 and 2020-2021 mountain ranges. (Note that in all the plots that follow, observed or actual data are shown in blue and predicted or expected data are shown in green.)

As can be seen, the actual values in 2019-2020 and 2020-2021 are greater than the predicted values. They are, however, within the 95 percent confidence interval resulting from the uncertainties in the values of the slope and intercept of the best-fit (green) straight line. These uncertainty ranges are indicated by the vertical error bars. (Calculating the uncertainties and confidence interval is explained in detail at the link provided above.)

Accordingly, the numbers of actual deaths in the two pandemic mountain ranges cannot be asserted to be statistically significantly different from the predictions. That the mountain range deaths in these two seasons are not anomalous can be seen in Figure 6. This charts the difference between the actual death count and the predicted number corresponding to the straight line.

Figure 6. The deviation of the number of deaths in the mountain ranges about the linear trend line shows a remarkably regular oscillation. This is consistent with flu strains of alternating severity and/or flu shots of alternating effectiveness. The deviation from the trend line for the pandemic years is clearly not excessive, being less than that in the 2014-2015 mountain range and no greater than the 2012-2013 mountain range. The black bars indicate range of uncertainty in predicted numbers at 95 percent confidence level. The two non-pandemic mountain ranges above the range of uncertainty are likely due to severe strains of flu or inappropriately formulated flu shots. (Chart by Jim Mason)

Also interesting is that the actual death count in the mountain ranges shows an almost regular annual oscillation about the trend line. This is unusual for random variables and suggests some sort of systematic variation, perhaps flu strains regularly oscillating between less and more virulent and/or flu shots being alternately more and less effective.

Most importantly, the difference between the number of deaths in the mountain ranges of both pandemic seasons and the predicted values for these seasons based on previous years, is not significantly different from the differences in other years. In particular, both are less than the difference observed for the 2014-2015 season (which exceeded the uncertainty range) and very nearly the same as observed in 2012-2013 and 2017-2018.

Seasonal Perspective ConclusionThe incremental number of deaths during the fall-winter-spring “flu” season of the pandemic years does not appear statistically significantly different from the number expected based on the numbers observed during corresponding pre-pandemic seasons.

Monthly Perspective – By Month Individually

This approach uses the death counts for each month in the pre-pandemic years to predict the expected count in the same month during the pandemic years. It is thus similar to the OWID analysis but using a monthly rather than weekly time interval. The number of deaths each month for the period 2010 through 2021 is shown in Figure 7.

Figure 7. Observed number of monthly deaths for each year since 2010 shows (i) the gradual increase in number of deaths over time, (ii) the reasonably consistent shape of the annual plot, (iii) the different shape of the 2020 plot with an obvious large peak in April and May, and (iv) a possible elevated region in the latter half of 2021. (Chart by Jim Mason)

Figure 7 shows the already discussed, gradual year-over-year increase. In addition, it can be seen that the shape of the graph across the year is more or less the same each year, with the exception of the two pandemic years. The first pandemic year, 2020, has an obvious peak in April-May and a slight elevation in the latter part of the year. The second pandemic year, 2021, has a reasonably typical shape during the first half of the year, a slightly elevated shape during the late summer and early fall, with a return to a typical shape towards year-end.

A detailed, month-by-month analysis over the time-frame 2010-2019 (essentially looking at the monthly data in vertical lines) was performed to determine the maximum time-frame for each month that could reliably be used to predict the expected deaths in that month for 2020 and 2021. This revealed that:

  1. The plains months of June-September had the same two-segment linear composition as the seasonal data. That is, for these months 2010-2012 formed a straight line with one slope and 2013-2109 formed another straight line with a different slope.
  2. The months December, January and February did not have this same pattern but rather showed a decided oscillatory pattern that extended over the entire 2010-2019 time-frame.
  3. The months March, April and May (which precede the plains months) and October and November (which follow the plains months) had a mix of these two constructs with the two-segment linear construct becoming increasingly obvious from March to May but less obvious from October to November.

Based on this, the expected number of deaths in each month of the two pandemic years was estimated as follows:

  1. For December, January and February, a linear extrapolation of the entire 2010-2019 time-frame.
  2. For other months, a linear extrapolation of the period 2013-2019.
Figure 8. The observed and predicted number of deaths for the months January, February, March and April over the study period, together with the range of uncertainty in predicted numbers at the 95 percent confidence level. Observed numbers are within the uncertainty interval for all months/years except April 2020. (Chart by Jim Mason)
Figure 9. The observed and predicted number of deaths for the months May, June, July and August over the study period, together with the range of uncertainty in predicted numbers at the 95 percent confidence level. Observed numbers are within the uncertainty interval for all months of the two pandemic years except May 2020, which is significantly higher than the upper limit, and May, July and August 2021, which are marginally above the upper limit. Months in 2010 that are marginally above the upper limit may indicate anomalous behaviour, but this was not pursued; these months were not used for predicting numbers in these months of the pandemic years. (Chart by Jim Mason)
Figure 10. The observed and predicted number of deaths for the months September, October, November and December over the study period, together with the range of uncertainty in predicted numbers at the 95 percent confidence level. Observed numbers are within the uncertainty interval for all months of the pandemic years except November and December 2020, and September and October 2021. October 2010 is also above the upper uncertainly level, which may indicate anomalous behaviour, but 2010 was not used in making predictions for this month of the pandemic years. (Chart by Jim Mason)

In all months, the assumed underlying linear relationship between number of deaths in the month and the year provides a reasonable fit to the data used to define the relationship – i.e. 2010-2019 or 2013-2019. For all but December and January the goodness of fit (R2) value is ≥ 0.90 (1.00 indicating a perfect fit). This means that for those months, the linear model can be used reliably to predict the numbers of deaths. For December the goodness of fit is 0.77 and for January 0.66. This poorer fit is due to an almost periodic oscillation about the linear variation. Attempts to model this with a sine wave oscillation were unproductive. This means that for these two months, the linear prediction is less reliable and results in larger uncertainty ranges, as shown by the vertical bars in the charts.

Figures 11 and 12 graph the actual and predicted monthly deaths (together with the uncertainties in the predicted values) by month for 2020 and 2021. The actual values for most months are within, or very nearly so, the uncertainty ranges about the predicted values. For the pandemic years, the notable exceptions are:

  1. 2020: April and May, which are well above the predicted range, and November and December, which are marginally above the predicted range.
  2. 2021: September and October, which are somewhat above the predicted range, and July and August, which are less so.

Several months in 2010 are also slightly higher than the upper level of the uncertainty range, suggesting something anomalous occurred that year. Because the focus of this analysis is on the pandemic years, this was not pursued. The 2010 data are included in the dataset for defining the straight line used for estimating the expected deaths during the pandemic years only for months December, January and February. In these months, the 2010 values are well within the uncertainty limits; accordingly, the anomalous 2010 values in other months have no impact on this analysis.

Figure 11. The observed and predicted number of deaths for each month in 2020 together with the range of uncertainly in predicted numbers at the 95 percent confidence level. April, May, November and December are above the uncertainty range; January, February and March are below the predicted numbers but within the uncertainty range. (Chart by Jim Mason)
Figure 12. The observed and predicted number of deaths for each month in 2021 together with the range of uncertainly in predicted numbers at the 95 percent confidence level. May, July, August, September and October are marginally above the uncertainty range, while March and April are below the predicted numbers but within the uncertainty range. (Chart by Jim Mason)

In 2020, the higher-than-predicted numbers of deaths in April and May are clearly visible. They are above the upper end of the uncertainty range and, so, can be said to be higher than expected with a very high degree of confidence. November and December also have higher-than-predicted numbers (although less so than for April and May), are also above the upper limit of the uncertainty and so can be concluded to be different from the predicted numbers with only a 5 percent chance of being wrong. It is therefore reasonable to conclude that there were excess deaths in these months.

Actual numbers for most months in 2021 are within the uncertainty ranges of the predicted numbers. July and August are slightly outside the range of uncertainty for predicted numbers. These could be instances of the 5 percent of observations that fall in this region without being anomalous or could, in fact, be statistically significantly different, representing a small number of excess deaths. September and October are clearly outside the range of uncertainty and, accordingly, can be said to be higher than expected with a very high degree of confidence suggesting excess deaths in those months..

Monthly Perspective – By Month Cumulatively

For both 2020 and 2021, some months are higher than predicted, some months are lower, and the magnitude of uncertainty in some of the months is quite large. It is thus not obvious whether the total in any month or for each entire year is statistically significantly different from what would be predicted based on pre-pandemic history.

One way to assess this is by looking at the cumulative deaths to the end of each month during the year compared to the predicted values. The two values are charted in Figures 13 and 14 for each pandemic year. The scale makes it difficult to see any differences, so Figures 15 and 16 plot the cumulative difference between the observed number and the predicted number together with the range of uncertainty in the difference resulting from the range of uncertainty in the predicted number. For reference, the zero line, representing perfect agreement, is shown in green. From these, it would appear that, by the end of the year, 2020 experienced significantly more deaths than predicted, but 2021 did not.

Figure 13. The observed and predicted cumulative number of deaths to the end of each month in 2020 together with the range of uncertainty in predicted numbers at the 95 percent confidence level. The observed total deviates significantly from the predicted as of May. See Figure 15 for finer detail. (Chart by Jim Mason)
Figure 14. The observed and predicted cumulative number of deaths to the end of each month in 2021 together with the range of uncertainty in predicted numbers at the 95 percent confidence level. The observed total deviates from the predicted as of September. See Figure 16 for finer detail. (Chart by Jim Mason)
Figure 15. The difference between the observed and predicted number of deaths accumulated to the end of each month of 2020, together with the range of uncertainty at the 95 percent confidence level, based on the predicted numbers. For any month in which 0 falls outside the uncertainly interval (black bar), the observed number of deaths is considered significantly different from the predicted number. The April-May anomaly makes the observed cumulative number of deaths statistically greater than the predicted number at the end of May. This is followed by further increases towards the end of the year. (Chart by Jim Mason)
Figure 16. The difference between the observed and predicted number of deaths accumulated to the end of each month of 2021, together with the range of uncertainty based on the predicted numbers. For any month in which 0 falls within the uncertainly interval (black bar), the observed number of deaths is not considered significantly different from the predicted number. This is clearly the case in all months. (Chart by Jim Mason)

Monthly Perspective ConclusionComparing actual monthly deaths during the pandemic years to predicted monthly deaths using pre-pandemic monthly deaths yields mixed results. For some months actual deaths are greater than the predicted; for others, less. Actual deaths in April and May 2020 (the first wave in Canada; scroll to Figure 2 at the link, then select “deaths”) were substantially larger than the predicted numbers for these months. In other months, while the actuals were greater than the predicted, this was not dramatically so. And, as noted, in some months, the actuals were less than predicted.

Summing the monthly numbers through the year leads to different results for 2020 and 2021. In 2020, total actual deaths start out below predicted deaths in January-March. Large numbers of deaths follow in April and May, jumping the total cumulative number of actual deaths to about 4 percent higher than the total predicted deaths. The gap continues to increase slowly, and reaches a total of 12,500 at year-end. Importantly, the uncertainty interval in the predicted value means that total excess deaths for 2020 could be as few as 6,700, or as many as 18,300.

In 2021, the ratio also starts out negative and remains so until mid-year, then moves into the positive, reaching a cumulative total in higher-than-predicted deaths of about 5,000 in November. Notably, the value is always less than the uncertainty in the difference. Thus, excess deaths in 2021 cannot be reliably asserted to be greater than zero.

The individual monthly and cumulative monthly perspectives are somewhat at odds with the conclusion in the seasonal analysis that there were no excess deaths in either pandemic year. This is because the opening part of the 2019-2020 mountain range, which does not affect the 2020 numbers, is lower than typical/predicted (which could indicate a milder-than-normal flu season). In the seasonal analysis this offsets the larger-than-predicted numbers in April and May 2020 (which also drive the 2020 total annual numbers, discussed in the following section). Conversely, the higher-than-typical fall months of 2020, while affecting the 2020 annual numbers, are offset in the seasonal analysis by the lower winter-spring numbers for 2021.

Annual Perspective

The fourth and final way that this analysis examined the data was to look at the total number of deaths each year (ignoring seasonal and monthly data) and to use actual annual data from the pre-pandemic years to predict the expected number of deaths during the pandemic years. Since the seasonal analysis showed a distinct difference between periods 2010-2012 and 2013 onward, only the years 2013 through 2019 were used as the predictive years and prediction was done by extrapolation of a linear fit to these data. The result is shown in Figure 17 together with the uncertainty range for each annual prediction that results from the goodness of fit of the linear regression.

Figure 17. The annual observed and predicted number of deaths for 2010 through 2021, together with the range of uncertainly at the 95 percent confidence level. Years when the observed number falls within the uncertainly range are considered not significantly different from the predicted number. The exception is 2020, which has a significantly higher number of observed deaths than predicted. (Chart by Jim Mason)

The linear model fits the 2013-2019 data quite well (R2=0.97). Extrapolating the resulting relationship through 2020 and 2021 results in the observed deaths in 2020 being above the upper limit of the uncertainty interval but those in 2021 being below the upper limit. These results are consistent with the monthly cumulative analysis.

The April-May 2020 Anomaly – From the preceding analyses, it is clear that the number of deaths in April and May 2020 is a major anomaly that, by itself, determines the outcomes of the analysis for that year. This period corresponds to the crest of Covid-19’s first wave in Canada, governments’ imposition of “non-pharmaceutical interventions” – principally lockdowns – and the tragedies of forced isolation and thousands of deaths in seniors’ residences and long-term care facilities (LTCF), some of which may have been due to intentional population triage, “a deliberate effort to protect the healthcare system at the expense of vulnerable LTCF residents.”

Examining what the situation would likely have been without these two months proved instructive. That is, if instead of imposing the aforementioned measures, Canada’s governments had followed established pandemic management and/or emergency management techniques, or had taken note of clearly observable events in other countries – that the elderly, especially those with other underlying health conditions, were most vulnerable while the young and healthy were far less so. Had Canada’s governments implemented a more suitable and better targeted set of measures, and had deaths been successfully held to the levels expected from historical data, what would the overall situation have been?

This can be seen by simply setting the number of deaths for these two months equal to the number predicted from the historical data.

Figure 18. Difference between actual number of deaths, adjusted to eliminate the impact of the April-May anomaly, and predicted number of deaths, accumulated to the end of each month of 2020, together with the range of uncertainty in predicted numbers at the 95 percent confidence level. Zero is within the uncertainty interval for each month, which means that none of the monthly differences can be considered significantly different from zero. (Chart by Jim Mason)
Figure 19. The observed annual number of deaths by year, with the April-May anomaly removed from the 2020 number, together with the range of uncertainty in predicted numbers at the 95 percent confidence level. Since all observed numbers are within the uncertainty interval, it is apparent that only the April-May anomaly caused 2020’s observed deaths to be significantly higher than the number to be expected from the historical data. (Chart by Jim Mason)

As can be seen from these two figures, without the April-May 2020 anomaly:

  1. The cumulative number of deaths at the end of each month during 2020 is not statistically significantly different from the number predicted.
  2. The total number of annual deaths in 2020 is not statistically significantly different from the number predicted.

Making Sense of it All

As previously noted, this analysis was undertaken to examine whether or not there were excess deaths during the pandemic and, if so, whether these deaths can be unequivocally attributed to the virus or might have been (partially or wholly) the consequence of governments’ responses to the virus. The results can be described as more than indicative but less than definitive, i.e., the conclusion concerning excess deaths depends on the time-frame used.

To recap briefly, this analysis used four different time-based approaches in attempting to illuminate what happened, discern patterns and reach conclusions about excess deaths: seasonal, individual monthly, cumulative monthly and annual.

Looking at the number of deaths on a seasonal (October through May) basis, in neither 2019-2020 nor 2020-2021 were the numbers of actual deaths statistically significantly different from predicted/expected numbers.

Looking at each month separately, results were mixed. In some months the actual number of deaths was less than the predicted number. In some months it was more than predicted but not significantly so statistically, as actual numbers were within the uncertainly range of the predicted numbers. In some months it was significantly more, but only marginally greater than the upper limit for the uncertainty range.

In only two months – April and May 2020 – were the numbers of observed deaths dramatically above the upper limit of the uncertainty interval. These results are qualitatively the same as those produced by StatsCan and OWID on a weekly basis; that is, the April-May period is clearly anomalous while the other months are either not unusual or only marginally so.

Looking at the cumulative number of deaths to the end of each month during the year, 2020 actual deaths were not significantly different from predicted deaths until after May, after which they showed a slight upward trend to the end of the year. Actual cumulative monthly deaths in 2021 were not significantly different from predicted deaths through year-end. Again, the April-May period is sufficiently anomalous to determine the result for 2020.

Looking at the total number of deaths for each year, 2020 actual deaths were above the upper limit of the uncertainty range in the predicted value, but those in 2021 were not. The 2020 result is entirely determined by the number of deaths in April and May.

In summary, April and May 2020 form a significant anomaly that determined the outcomes of the individual monthly, cumulative monthly and annual analyses for 2020. This corresponds to the crest of the pandemic’s first wave in Canada, when LTCFs and other places of concentrated seniors’ residency were very hard hit by deaths.

Absent this anomaly, the numbers of cumulative monthly deaths and total annual deaths for 2020 and 2021 are not, statistically, significantly different from what would be expected from pre-pandemic data. These months do not have the same impact on the seasonal analysis because of lower-than-predicted numbers during the preceding months of the 2019-2020 season, especially the first three months of 2020.

The Grim Reaper vs. the Government

The pandemic does not appear to have resulted in a significant increase in deaths in Canada based on historical data from the preceding decade, except for April and May 2020, when increases were dramatic enough to determine the annual outcome for 2020 but not for the 2019-2020 flu season.

The absence of a significant uptick in deaths in 2020 and 2021 except for the April-May 2020 anomaly could be interpreted as an indication that the government measures after the April-May 2020 time-frame worked perfectly to exactly counter the number of deaths that would have otherwise occurred, or as an indication that the government measures accomplished very little, since little was required.

The April-May 2020 anomaly, which coincided with the harshest and most widespread non-pharmaceutical interventions in nearly all Canadian jurisdictions, suggests that these measures were largely ineffective at preventing deaths, particularly in the most vulnerable population segments. Accordingly, the likelihood of the measures following this period being so finely tuned as to achieve balance is considered quite small, especially since the measures varied dramatically from time to time and province to province.

Based on this analysis, it is not unreasonable to conclude that government measures had little impact on the overall, Canada-wide number of deaths during 2020 and 2021, except for their failure to prevent the deaths during the first wave, particularly in LTCFs and other places of concentrated seniors’ residency. It is also possible that a different government response to the first wave that focused more on the already-known particularly vulnerable demographics might have ameliorated or avoided the steep death toll of April-May 2020. Certainly, this is the view eloquently expressed by David Redman, former Executive Director of the Alberta Emergency Management Agency.

The nationwide April-May 2020 lockdowns failed abysmally at preventing thousands of deaths from Covid-19. Yet even after the first wave subsided, these variable measures – travel restrictions, bans on purchasing “non-essential” items, store capacity limitations, restaurant closures, prohibition of many common activities, to name a few – remained in place. (Sources of photos: (top) Shutterstock; (bottom left) CTV News Winnipeg; (bottom right) Evan Saunders)

It should also be said that since the government measures swept up nearly the entire population to varying degrees and differing over time, including segments at moderate, low or even immaterial risk from Covid-19, it is highly likely that the measures themselves caused a number of deaths. Accordingly, apparent excess deaths during the pandemic could have resulted from ill-considered and ineffective government measures rather than the virus itself.

Can the Contribution of Government Measures be Estimated?

One approach to assessing the relative contribution of government measures to excess deaths is provided by a different analysis of excess deaths reported in the Canadian Medical Association Journal (CMAJ). Entitled Excess mortality, COVID-19 and health care systems in Canada, the paper appears to show quite clearly that government measures contributed to the excess deaths, in many cases more so than the virus.

The CMAJ analysis looked at the correlation of the StatsCan weekly excess death data with weekly reported Covid-19 deaths on a province-by-province basis. To facilitate comparison among provinces, it used mortality rates (deaths per million population) rather than absolute numbers.

Excess deaths are those deaths above the number expected based on pre-pandemic experience. The factors present during the pandemic that were absent pre-pandemic that could account for these deaths are (i) the Covid-19 virus, and (ii) various government response measures. (Voluntary individual responses, whether rational or not – things like choosing to order groceries for home delivery, not travelling, drinking far too much alcohol, or deferring hospital visits to “help the system” – are grouped along with government measures since this sort of behaviour was largely a direct consequence of government measures or was encouraged by government or other public health-related communications.) Thus, there are two components of excess deaths during the pandemic: those that were caused by Covid-19 and those that were not caused by Covid-19 but, rather, the measures imposed by governments in attempting to contain Covid-19.

These two factors could result in three possible situations and various results, as illustrated in Figure 20.

Figure 20. Simplified dynamics among main causal factors during the pandemic showing the ways different excess death results might occur. (Chart by Jim Mason)

In the first situation, expected deaths and observed deaths are equal, resulting in no excess deaths. This could be because the pandemic was a non-event and neither the virus nor the government responses produced incremental deaths (situation 1A in Figure 20). It could also result from the virus causing some incremental deaths but the government measures averting an equal number by, for example, compelling people to stay off the roads and thus reducing traffic fatalities (situation 1B).

In the second situation, observed deaths are less than expected deaths. This could be because the virus caused no deaths, either because it was not particularly virulent or because the government measures prevented them, combined with government measures averting a number of deaths from other causes (situation 2A ). Alternatively, the virus could have caused some deaths but government measures averted a larger number of deaths (situation 2B).

In the third situation, observed deaths exceed projected deaths – there are, in fact, excess deaths. This could result from any of three possibilities. In the first (situation 3A), excess deaths are equal to Covid-19 deaths because there are Covid-19 deaths that government measures failed to prevent while government measures also failed to avert deaths from other causes. In the second (situation 3B), excess deaths are less than Covid-19 deaths because while government measures failed to prevent all Covid-19 deaths, they did avert some deaths from other causes. In the third (situation 3C), the number of excess deaths exceeds the number of Covid-19 deaths.

In the case of situation 3C, government measures were not only unable to prevent many Covid-19 deaths but were themselves causing deaths. If the number of excess deaths is more than twice the number of Covid-19 deaths, then government measures caused more deaths than the virus. This is conceptually illustrated in Figure 20.

So what actually happened? Figure 21 reproduces a graph from the CMAJ paper that provides the real-life application in British Columbia of the situations outlined above. This is overlain with graphic additions to indicate the occurrences of the various situations described above.

Figure 21. Excess deaths and Covid-19 deaths for British Columbia, with the corresponding situations illustrated in Figure 20 indicated with colour coding. During the long red intervals, government measures are apparently increasing the number of excess deaths over and above those resulting from Covid-19. (Source of chart: Kimberlyn McGrail, PhD, Excess mortality, COVID-19 and health care systems in Canada, CMAJ 2022 May 30; 194:E741-5. doi: 10.1503/cmaj.220337, Figure 1A. Annotation pertaining to Situations 1, 2 and 3, as described in the text, overlaid by Jim Mason)

The graph shows only two brief instances of situation 1, where excess deaths equal zero. These lie on either side of the also very brief situation-2 interval (shown in blue and annotated with a label), where excess deaths are less than zero. Since Covid-19 deaths exceeded 0 throughout this period, these intervals show instances of situation 1B and situation 2B, indicating that government measures briefly resulted in the avoidance of a sufficient number of deaths in other areas to offset the number of reported Covid-19 deaths.

Situation 3 prevailed for the remainder of the 18-month period depicted. That is, the number of excess deaths in B.C. was greater than zero throughout almost all of this period. Instances of situation 3A, where excess deaths equal Covid-19 deaths, are identified by the green vertical lines. These are singular points rather than periods, where the condition exists simply because of the trajectories of excess deaths and Covid-19 deaths, often headed in opposite directions, and can hardly be attributed to government measures.

Instances of situation 3B, where excess deaths are less than Covid-19 deaths, are indicated by the four orange bars. In these regions, government measures have not prevented all Covid-19 deaths (although they may have prevented some) but have somehow averted more deaths from other causes than were caused by Covid-19. The three brief instances and one period of situation 3B span approximately 6-8 weeks of the 18-month study period.

Situation 3C, where excess deaths exceed Covid-19 deaths, is indicated by red bars and, significantly, covers an astounding 91 percent of the 18-month period (80 weeks out of 88). (Note the extraordinary number of excess deaths during the approximately four weeks of the heat dome in July 2021. As illustrated by the lines in the figure, however, these were in addition to an upward sloping trend in excess deaths that had already been ongoing for about three months and continued for several months following the heat dome.) Since Covid-19 deaths do not account for all the excess deaths, it would seem that government measures (or population behaviour induced by government measures or, in the case of the heat dome, unusual events) are actually causing deaths.

In some periods, the non-Covid-19 contribution to excess deaths is quite large. In August 2020, Covid-19 deaths in B.C. were very low at about 1 per million, whereas total excess deaths were circa 15 per million, suggesting that the contribution from government measures was as much as 14 deaths per million population – 14 times that of Covid-19 itself. In these situations, it is not unreasonable to state that the government’s response to the viral pandemic was worse than the virus itself.

All other provincial profiles presented in the CMAJ paper also show significant periods when excess deaths were greater than Covid-19 deaths, except for Quebec, where excess deaths were consistently less than Covid-19 deaths.

The CMAJ paper also conducted a cumulative analysis similar to the monthly cumulative one done for this essay, looking at the total excess deaths and Covid-19 deaths for each province over the entire timeframe (reproduced in Figure 22).

Figure 22. The relationship of excess deaths to reported Covid-19 deaths in each Canadian province from March 2020 through October 2021. In six provinces representing the large majority of Canada’s population, excess deaths exceeded reported Covid-19 deaths – in some cases by a factor of two, three or even greater – indicating that government measures and population responses to the pandemic, many of which were a direct outcome of government policies, were causing more deaths than Covid-19 itself. (Source of chart: Kimberlyn McGrail, PhD, Excess mortality, COVID-19 and health care systems in Canada, CMAJ 2022 May 30;194:E741-5. doi: 10.1503/cmaj.220337, Figure 2)

For most provinces, the cumulative number of excess deaths exceeds the number of Covid-19 deaths, suggesting that government measures (and population responses) themselves caused some excess deaths. In several provinces, excess deaths are more than twice as great as Covid-19 deaths, suggesting that deaths caused by government measures exceeded the deaths caused by the virus. Among the exceptions, the most striking and important is Quebec, where excess deaths were significantly less than Covid-19 deaths. Given that Quebec had about twice the reported Covid-19 mortality rate of any other province, this would indicate Quebec’s government measures were highly ineffective at curbing Covid-19 deaths but highly effective at avoiding deaths from other causes.

Conclusion

This analysis demonstrates that the official monthly death data published by Statistics Canada do not indicate that Canada-wide deaths during 2020 and 2021 were substantially higher than would be expected based on data from the previous decade, using accepted methods of projection and uncertainty ranges, with two monthly exceptions. These were April and May 2020, during which the number of actual deaths was well above the upper uncertainty limit of the expected number. The combined number of deaths in April and May 2020 was so large as to cause the total number of deaths in 2020 to exceed the upper limit of what might have been expected based on the previous decade.

The difference between the observed and projected number of total deaths for 2020 – the estimated number of excess deaths – is calculated to be 12,495, with upper and lower figures, based on the uncertainty interval for 95 percent confidence, of 18,280 and 6,710. The preponderance of these deaths was of residents at LTCHs and other places of concentrated seniors’ residency and, as noted earlier, may have stemmed in part from intentional population triage aimed at “saving” the healthcare system from overloading. Even so, the calculated excess death number is still not so large as to make the number of deaths during the October 2019 to May 2020 flu season anomalously large. For 2021, excess deaths cannot be attributed with 95 percent confidence, because the reported death figures fall within the uncertainty range of what would be expected that year.

On its face, these results would seem to suggest that government measures imposed in response to Covid-19 were effective at containing deaths. Since these measures failed calamitously during the April-May 2020 anomaly, however – the very time when they were most needed and most vocally promoted by governments – and subsequently varied widely from province to province and from time to time, it is not unreasonable to conclude that these measures were only accidentally effective on an aggregate basis.

This conclusion of accidental, if any, effectiveness is consistent with the data presented in the CMAJ paper, discussed above. It quite markedly indicates that for most of the time in most provinces, government measures contributed to the excess death count. In plain language, the government response to Covid-19 got people killed. The notable exception was Quebec, which had the highest reported Covid-19 mortality rate but among the lowest excess death rates. When viewed from a whole-country perspective, Quebec’s relatively large population offset the other provinces’ higher excess mortality rates and lower Covid-19 mortality rates.

Think twice before destroying lives next time: With excess deaths exceeding Covid-19 deaths in the majority of Canadian provinces, government overreach and collateral damage were likely more devastating than the virus itself. This should encourage profound reflection about emergency management before the next major national crisis.

All these data strongly suggest that it is incumbent upon governments to give far more consideration to the collateral damage that may be caused by measures imposed to deal with perceived emergencies. While this essay has considered only deaths resulting from these measures during the Covid-19 pandemic, their broader economic and quality-of-life impacts should not be summarily dismissed, as they can have a longer-term impact on individual health, especially mental health, as well as longevity and mortality. In short, governments need to be much more careful when playing with the lives of millions of Canadians.

Appendix A – Other Excess Death Analyses

Statistics Canada – StatsCan publishes a weekly updated but lagging estimate of excess deaths. It uses a Poisson probability distribution modified by a number of ad hoc additions. The Poisson distribution is a discrete probability distribution, i.e., it provides the likelihood of observing a particular, discrete, integer number. It expresses the probability of a given number of events occurring in a fixed interval of time (or space) if these events occur with a known constant mean rate and independently of the time since the last event.

On the surface, this would seem to be a good model for predicting the number of deaths occurring in any given fixed interval – day, week, month or year. There are, however, some complications. The mean rate of deaths (per unit of time) is not constant over time. As discussed in the article, it varies by year (an increasing and aging population will lead to more deaths), by month (deaths during the winter “flu season” are higher than during the summer), by week, and possibly even by hour (more deaths occur at night than in daytime). Because of this, estimating expected deaths using the Poisson approach requires a number of modifications.

As per the StatsCan website, its approach is “based on a quasi-Poisson regression model fit to weekly…death counts (all-cause) spanning a selected reference period of approximately four years (2016-2019)…An overdispersed Poisson generalized linear model with a linear time trend and a seasonal factor is fit to the data. The seasonal component aims to represent the expected pattern across weeks that repeats from year-to-year, and consists of a zero-order spline term with 11 knots, representing 10 distinct periods within a given year. The 10 periods are split between a single 7-week period corresponding to the current week being estimated and the 3 preceding and subsequent weeks, and 9 other 5-week periods corresponding to the rest of the year.”

So the StatsCan analysts performed a number of ad hoc modifications to the basic Poisson distribution and, as a result, there are numerous parameters used to “tune” the model’s performance. Consequently, it is rather complex and abstruse. StatsCan acknowledges its approach suffers from a number of drawbacks:

  • “Analysis of death by date (or week) of death is inevitably distorted by delays in reporting”; and
  • “Reporting delays are susceptible to change over time.”

Various “adjustment factors” are used to account for these.

StatsCan provides an estimate of uncertainty in its estimate of predicted deaths, using the standard 95 percent confidence interval for assessing the significance of differences between the predicted number and actual number. That is, if the difference exceeds the interval’s upper limit, the number of actual deaths can be considered statistically significant – a real excess death count – with only a 5 percent chance of the conclusion being wrong. The upper limit of this uncertainly interval is shown on the graphs in the StatsCan website. In the author’s opinion, the complexity of the StatsCan approach makes it difficult to assess the accuracy of its results and, hence, reduces its value.

Our-World-In-Data (OWID) – This organization’s website includes Canada in its analysis of excess deaths. OWID performs the analysis on a weekly basis throughout a calendar year. The number of projected deaths is “an estimate produced by Ariel Karlinsky and Dmitry Kobak as part of their World Mortality Dataset (WMD). To produce this estimate, they first fit a [linear] regression model for each region using historical deaths data from 2015–2019 (or as many years from this period as are available). They then use the model to project the number of deaths we might normally have expected in 2020-2022. Their model can capture both seasonal variation and year-to-year trends in mortality.”

This approach does not assume any underlying statistical model for deaths but simply treats each week separately, estimating the projected number of deaths for that week by fitting a straight line through the same week of the five years 2015-2019. The number of expected deaths for the weeks during the pandemic years is then estimated by extrapolating the straight line. OWID also does an uncertainty calculation for the projected number of deaths using standard statistical analysis techniques.

Jim Mason earned a BSc in engineering physics and a PhD in experimental nuclear physics. His doctoral research and much of his career involved extensive analysis of “noisy” data to extract useful information, which was then further analyzed to identify meaningful relationships indicative of underlying causes. He is currently retired and living near Lakefield, Ontario.

Source of main image: Shutterstock.

Love C2C Journal? Here's how you can help us grow.

More for you

They Would Call Me a “Denier” – Let Me Explain what I Believe about Residential Schools in Canada

In our largely “post-truth” society, the validity of a given statement is increasingly assessed based on who is making it. There are even those who believe that only some should be allowed to say certain things – while others should be scorned or even imprisoned for uttering the same words. This increasingly describes the discursive landscape concerning Canada’s Indian Residential Schools and whether Indigenous children disappeared from and/or were murdered there. Drawing on his lived experience as a onetime residential school employee, on his long academic record and, not least, on his personal courage in the face of those who wish to criminalize “denialism”, Rodney Clifton presents a humbly argued plea for Canadians to judge their country’s residential school record according to the truth – the actual, factual truth.

One Job We Can’t Let AI Replace: Philosopher. Rethinking the Ethics of Artificial Intelligence

We stand on the precipice of a new technological age. Artificial intelligence promises (or threatens) to upend every aspect of modern life – from employment to entertainment, manufacturing to warfare – as well as the very relationship between humanity and the machinery it creates. Given AI’s potentially cataclysmic consequences, D.C.C. Randell argues it is imperative that we set not merely regulatory and technical boundaries around its development, but ethical ones as well. Combining the warnings of current AI experts with the wisdom of philosophers and moralists from past ages, Randell explains the dangers posed by allowing the AI revolution to continue unfettered and proposes steps to bring it in line.

Arresting Developments: The Persecution of David Menzies and the Fight for Media Freedom in Canada

The sight of a journalist getting roughed up and hustled off by police as he tries to ask a question of a public figure in a public place is one you might expect to witness in a banana republic or present-day Russia. But it has happened four times in Canada just this year to veteran journalist David Menzies of Rebel News. Menzies is an old-fashioned street journalist – right down to his trademark fedora – asking straight questions and digging for the truth no matter the consequences. In this instalment of C2C’s Courageous Canadians series, Associate Editor Brock Eldon sits down with Menzies to talk about his run-ins with the law, his determination to publish without fear or favour, and the state of Canadian journalism.

More from this author

Truths the Emergencies Act Inquiry Must Reveal 

The Freedom Convoy became not just a conveyance to bring protesters to Ottawa but, for whatever reason, a kind of magnet for lies. Falsehoods, misdirection and dissembling proliferated about everything from alleged racism and violence to whether or not downtown Ottawa was actually “occupied” or “blockaded” to the technicalities of who even wanted the Emergencies Act and who actually ordered the freezing of bank accounts. Now, writes Jim Mason, it’s time for the truth – all of it, every material piece. In Mason’s view, the commission of inquiry needs to look back well beyond the Convoy’s arrival and assess decisions and claims by the Trudeau government that, Mason argues, set the conditions for the Convoy itself.

The Evaporating Effectiveness of Covid-19 Vaccines as Shown by the Newest Ontario Government Data

As Covid-19 spreads unchecked and becomes seemingly as common as colds and flus, it’s ever-harder to maintain confidence in the mRNA vaccines developed in 2020 to combat the virus’s original strain. Yet the prevailing view still seems to be, Where’s the harm? Might as well be extra-safe and get that booster (or even second booster). But what if an objective statistical analysis of official government data showed not only that the currently available vaccines do nothing to prevent infection by Omicron and its subvariants, but actually make people more susceptible? Following on his groundbreaking analysis of Ontario infection case data in February, Jim Mason returns to the task with this evaluation of the province’s most up-to-date data. Hint: the vaccines aren’t getting any better.

Do Epidemiological Data Support the Case for Vaccine Mandates?

“Follow the science!” seems to have largely given way to “Vaccinate! Obey! Or else!” the latter threat incorporating everything up to and including martial law. Perhaps there’s no better time, then, to take a look at what the actual data say and where they might lead – if we choose to listen. In this exclusive research-based essay, Jim Mason lays out and meticulously examines the most recent three months’ worth of Ontario government Covid-19 case data. Applying accepted epidemiological formulae in his analysis, Mason vividly charts the vaccines’ increasingly worrisome performance throughout the Omicron wave.

Share This Story

Donate

Subscribe to the C2C Weekly
It's Free!

* indicates required
Interests
By providing your email you consent to receive news and updates from C2C Journal. You may unsubscribe at any time.