Empirical Analysis I: What is the impact on waiting time of increased government health spending?

[Previous] [Contents] [Next]

Data

Because of these various possible connections between spending and waiting, empirical analysis is vital. Examination of the nature of the link between government health spending and waiting time is made possible by combining two primary data sources, the Canadian Institute for Health Information (CIHI) National Health Expenditure Database, and The Fraser Institute's annual measurement of waiting time. Both CIHI and The Fraser Institute report their data at the provincial level. The CIHI data covers the period 1975-1999, while The Fraser Institute has collected waiting time data for the period 1990-1998. Prior to 1993, however, the Institute's survey did not include oncologists, either medical or radiation. Hence, the median waiting time, averaged across all reported specialties, only began including oncology waiting times in 1993. Therefore, the analysis contained in this study pertains to the period 1993-1998 (with a few exceptions).

To understand this analysis better, it is important to describe the nature of these two main data sources in some detail. Consider first the waiting time data. This data has been collected by means of a survey of Canadian specialist physicians. Early years of the survey did not all include all provinces, or all specialties (e.g., oncology, as noted earlier). Beginning with the 1993 data, the members of 12 different specialties in all 10 provinces were surveyed. Those 12 specialties are: cardiovascular surgery, general surgery, gynaecology, internal medicine, medical oncology, neurology, ophthalmology, orthopaedic surgery, otolaryngology, plastic surgery, radiation oncology, and urology.

In each of the annual surveys, questionnaires were mailed to every member of the Canadian Medical Association practising in those fields. Thus, the survey population has ranged from over 7,500 in 1993 to over 9,300 in 1998. From this pool of potential respondents, the number of respondents has ranged from over 2,200 to almost 2,700, with response rates varying from 23 to 31 percent.

Each respondent reports two types of waiting times: how long a "new patient" would have to wait for a "routine office consultation" and after that consultation, for a variety of treatments performed by that type of specialist.11 For both the first segment of waiting - waiting for consultation - and the second segment - waiting for treatment after consultation - the median value of the specialists' responses is calculated. Then, to determine a single value to reflect the second segment of waiting, the median waiting times for the various treatments offered within a particular specialty are averaged (with the average weighting for each type of treatment being the frequency with which it is performed within that specialty in that province). This process then allows a number to be reported as the waiting time for treatment for "neurology," for example. Finally, this second segment of waiting can be added to the first segment to obtain the total median waiting time between referral from a general practitioner to a specialist and actual treatment.

As a result of this procedure, median waiting times for each of the 12 medical specialties are calculated for each province. To derive an overall median waiting time for each province, the median waits for each specialty are averaged together. This average is weighted, and the weights assigned to each specialty waiting time are determined by the proportion of all specialists in that province belonging to each of the specialty groups. Hence, if 10 percent of all specialists in Manitoba are general surgeons, then the weighting factor by which the Manitoba general surgery waiting time is multiplied is 0.1.

The total waiting times for 1993-1998, computed by this method, are listed in table 2. The data ranges from a minimum value of 7.4 weeks (Quebec, 1994) to a maximum value of 20.2 weeks (Saskatchewan, 1998). The simple average (across provinces) ranges from a low of 9.95 weeks in 1994 to a high of 14.21 weeks in 1998.

This data was paired with provincial data, from the same time period, measuring government spending on health (Canadian Institute for Health Information, 1999b). This data was collected and compiled by the Canadian Institute for Health Information (CIHI). As noted previously, the CIHI data series extends back to 1975. For each year, CIHI collected federal and provincial figures for spending on 7 categories intended to capture all uses of health spending: hospitals, other institutions, physicians, other practitioners, drugs, capital, and "other."12

Although the CIHI database also contains analogous information on private health spending, this study largely focuses on government spending and its impact. The main reason for this emphasis is that government is the primary health spender (69.6 percent in 1999) in Canada, and is the almost exclusive provider of the treatments covered by The Fraser Institute's waiting time survey. As well, this focus corresponds directly to The Fraser Institute's traditional concentration on the impacts of government policies and activities.

In order to make meaningful comparisons among provinces, spending was calculated on a per capita basis, and was further adjusted to incorporate differences in medical costs over time and across provinces. This adjustment relied on the health care component of the Consumer Price Index, which was reported on a province-specific basis in the CIHI database (Canadian Institute for Health Information, 1999b).

The real per capita spending figures generated by this process are reported in table 3. During this period, average real per capita government health spending fell until 1996, and then rose rapidly between 1996 and 1998. The lowest spender during this interval was Newfoundland ($1,594 in 1993); the highest spender was British Columbia ($2,183 in 1993).

The factors (in addition to spending) used to explain differences among provinces and across time in the waits experienced were demographic, economic, and political, and are listed in table 4.

Finally, real disposable income per capita (RDISPINCPC) was expected to reduce waiting time because residents of richer provinces would have been more likely to avail themselves of private care, either within or outside of Canada, shortening waiting time.13 Table 5 contains the mean, minimum, and maximum values for all variables, including the waiting time variable (WAIT) and the spending variable (RPCPUSP).

Comprehensive regression analysis of this data confirms the impression given by the simple correlation between spending and waiting: greater provincial spending, per capita, has no impact on waiting. The first indication of this is provided by the regression results in table 6.

The aspect of central importance in this, and the other regression results to be presented, is the effect of spending on waiting, as indicated by the coefficient on the spending variable, RPCPUSP. In table 6, the t-statistic for that spending variable is .78, but in order for spending to have a statistically significant effect on waiting (at the 10 percent level), its t-statistic would have to be larger than 1.67 (for a sample of this size).14 This means that there is no serious reason to think that spending has any effect on waiting. The coefficient itself is small (.0021), and its lack of statistical significance means that spending should be interpreted as having no effect on waiting.

The only variables which are statistically significant in this regression are those measuring age (OVER64) and time (TIME). OVER64 is found to be positive and significantly related to waiting; provinces with a higher proportion of elderly have longer waiting times. TIME is also positive and significant, meaning that even controlling for changes in spending and other factors over time, waiting times have grown longer over the sample period, 1993-1998.

It is possible that the insignificant connection between spending and waiting found thus far is actually concealing the fact that spending significantly reduces waiting. One potential problem might be that while there is no apparent connection when we look across provinces, spending may in fact reduce waiting when we look within provinces. In other words, when we examine a cross-section of provinces for a given year (say, 1998), we find that some have high spending and long waiting (British Columbia), some have high spending and short waiting (Ontario), some have low spending and long waiting (New Brunswick), and some have low spending and short waiting (Quebec). If different provinces fit each of these four types, then a cross-section comparison may not reveal any wait-reducing effect of spending. At the same time, it may still often be the case that when a particular province spends more in one year than in another, its waiting time will fall. But if the cross-section effect described above is dominant, a standard regression, as in table 6, will indicate that there is no connection between spending and waiting even if there is such a connection on a province-by-province basis.

To address this concern, one can rerun the regression in table 6 with the inclusion of provincial dummy variables. As there are 60 observations (6 years, 10 provinces), a particular provincial dummy - say, for Manitoba - takes on the value of 1 for each of the 6 Manitoba data points, and the value of 0 for the other 54 data points. Inclusion of dummies for every province allows us to determine if the lack of a significant relationship initially found between spending and waiting is the consequence of concealment due to provincial differences.

As it turns out, however, inclusion of provincial dummies has no serious impact on the relationship between spending and waiting; see table A-1.

Even after those provincial dummies are included, spending still fails to have a significant effect on waiting. On the other hand, once provincial differences are taken into account, age is no longer a significant determinant of waiting, while the presence of a Conservative premier reduces waiting time, and comes fairly close to the level of significance (i.e., t-statistic > 1.67).

A second problem which might conceal the true effect of spending on waiting is reverse causation. The regression analyses presented in tables 6 and A-1 are based on a model in which spending affects waiting, but not the reverse. However, it is conceivable that there is also reverse causation - that waiting affects spending. One circumstance which might generate such reverse causation is if governments spend more when waiting is expected to be higher. In that case, attempts to estimate waiting as a function of spending (as in tables 6 and A-1) might be biased by this reverse link. Specifically, if the effect of spending on waiting is estimated without accounting for potential reverse causation, it is possible that no significant effect would be found even when the true effect of spending on waiting is negative and significant. In other words, if we fail to account for the possibility that waiting increases spending, we may conceal the reality that spending reduces waiting.15

A standard technique to determine if there is reverse causation is the Hausman test (Berndt, 1991). According to this test, however, reverse causation is not a problem in the data used in this study. To further confirm this, estimates of the relationship between spending and waiting were made assuming that there was reverse causation (even though the Hausman test indicated otherwise). Even with a model which assumed the presence of reverse causation, the effect of spending on waiting was still not found to be significantly different from zero.16

The central finding of the analysis to this point is an apparently remarkable one - higher real per capita government health spending does nothing to reduce waiting time in the current Canadian context. Given this evidence that aggregate government health spending (per capita) is ineffectual in reducing waiting times, a natural next question is whether any particular form of government health spending does reduce waiting times.

Analysis of spending components

As noted earlier, the CIHI aggregate spending data is subdivided into 7 categories: hospitals, other institutions, physicians, other professionals, drugs, capital, and "other." Consequently, for the period of analysis considered in this study, 1993-1998, figures are available for each province for each year describing expenditures on each of those 7 categories. As with the prior analysis, these figures can be adjusted for population and for differences in health costs over time and across provinces, yielding real per capita government spending figures in each of the 7 categories.

Before reporting the effects of these different spending components, it is important to provide their definitions. According to the CIHI (CIHI, 2000), the 7 categories are defined as:

• hospitals ("institutions where patients are accommodated on the basis of medical need and are provided with continuing medical care and supporting diagnostic and therapeutic services," including "acute care, extended and chronic care, rehabilitation and convalescent care, psychiatric care, as well as nursing stations or outpost hospitals");
• other institutions ("residential care types of facilities (for the chronically ill or disabled, who reside at the institution more or less permanently)," including "homes for the aged (including nursing homes), facilities for persons with physical disabilities, developmental delays, psychiatric disabilities, alcohol and drug problems, and facilities for emotionally disturbed children," and excluding "[f]acilities solely of a custodial or domiciliary nature and facilities for transients or delinquents");
• MDs ("professional fees paid by provincial medical care insurance plans to physicians in private practice," as well as "[f]ees for services rendered in hospitals... when paid directly to physicians by the plans," and "other forms of professional incomes (salaries, sessional, capitation)");
• other professionals ("expenditures for the services of privately practising dentists, denturists, chiropractors, massage therapists, orthoptists, osteopaths, physiotherapists, podiatrists, psychologists, private duty nurses, and naturopaths," as well as "dental assistants and hygienists" and "optometrists and dispensing opticians," including the costs of "dental prostheses, including false teeth and laboratory charges for crowns and other dental appliances" and "expenditures for eyeglasses and contact lenses");
• drugs (prescription and non-prescription, the latter of which includes "over-the-counter drugs" as well as "personal health supplies" such as "oral hygiene products... diabetic test strips and... incontinence products");
• capital ("expenditures on construction, machinery and equipment of hospitals, clinics, first-aid stations, and residential care facilities");
• "other" ("expenditures on home care, medical transportation (ambulances), hearing aids, other appliances and prostheses, public health, prepayment administration, health research and miscellaneous health care," where "prepayment administration" means "expenditures related to the cost of providing health insurance programs by either government or private health insurance firms," "health research" means "expenditures for research activities designed to further knowledge of the determinants of health, health status or methods of providing health care, evaluation of health care delivery or of public health programs," not including "research carried out by hospitals or drug companies in the course of product development," which are included in either "hospitals" or "drugs").

Spending by these components for each year for each province is reported in table 7 and appendix tables A-2 and A-3. Largest among the 7 components over the period 1993-1998 was "hospitals," with a mean value of $852 per capita (1999 dollars), although the provincial averages for that period (table A-3) varied widely, from $686 in Saskatchewan to $931 in Newfoundland. Next largest was spending on MDs, averaging $323 over the sample period, with provincial averages ranging from $260 in Newfoundland to $451 in British Columbia. Other categories reflecting large government health spending are "other" (mean $280, ranging from means of $176 in Nova Scotia to $392 in Saskatchewan) and "other institutions" (mean $216, ranging from $136 in Quebec to $282 in British Columbia).

To analyze the consequences of shifting spending among these categories while holding aggregate per capita spending constant, a series of 7 regressions was run. Each of these contains 6 of the 7 percentage component variables, as well as all of the other control variables for demographics, income, politics, and strikes, as well as aggregate per capita spending.17 A typical example of these is found in table A-8. This particular regression omits the percentage spent on "other institutions." Therefore, it answers the question: holding aggregate per capita spending constant, does waiting time change if spending is shifted away from "other institutions" and into the other 6 components? The regression results indicate that the answer is "yes": specifically, shifting spending from "other institutions" to drugs reduces waiting time. Similar results are found in 4 of the other 6 regressions - shifting spending from hospitals, MDs, capital, and "other" into drugs will reduce waiting time (see tables A-9 through A-12). Conversely, shifting spending away from drugs and into every other component except "other professionals" increases waiting time (see table A-13). Connected to this is the finding (table A-14) that shifting spending from "other professionals" into drugs has no effect on waiting.

In the 6 of the 7 regressions in which "drugs" appears, the coefficient on that variable indicates the magnitude of the effect of shifting 1 percent of spending away from other components and into drugs. Depending on the component from which the increase in the percentage devoted to drugs is drawn, the size of the reduction in waiting time varies from a low of 1.18 weeks (shifting away from "other institutions," as reported in table A-8) to a high of 1.66 weeks (when the shift is away from "other," as shown in table A-12). Conversely, shifting 1 percent of spending away from drugs and into other components increases waiting by anywhere from 1.19 weeks (if the shift is into "other institutions") to 1.67 weeks (if the shift is into "other"); see table A-13.

Overall, these results indicate that even within the current Canadian health care system, waiting time can be reduced, without increasing spending, by shifting spending away from hospitals, other institutions, MDs, capital, and "other" and into drugs.18 This also means that existing government spending, on the margin, in those five areas, is misspent, if the goal is to reduce waiting time. At the same time, increases in aggregate per capita spending, as noted earlier, do not reduce waiting times: untargeted spending is ineffectual in improving waiting times, given the current structures and mechanisms of Canadian health care.

Analysis of spending and waiting times for specific specialties

Beyond the analysis of the effect of aggregate per capita spending and its various components on waiting time, it is also of interest to focus on the effect of spending increases on waiting times in specific specialties. Of perhaps greatest significance among the specialties are those in which timely interventions can prevent mortality and substantial morbidity as well as pain and suffering. Foremost in terms of this criterion are the cancer specialties, medical and radiation oncology, plus cardiovascular and orthopaedic surgeries.

Consequently, a portion of the analysis concentrated on the potential connection between spending (in total, and in components) on waiting for these four types of medical services. For three of these specialties - radiation oncology, medical oncology, and cardiovascular surgery - regression analysis indicated that not only did additional (aggregate) per capita spending not reduce waiting times, it appeared actually to increase waiting times. The only exception to this among the four was orthopaedic surgery; waiting time for it was not significantly affected by spending. These results are reported in tables A-15 to A-18.

One possible explanation for this statistical finding is reverse causation. Specifically, the fact that higher-spending provinces have longer medical oncology, radiation oncology, and cardiovascular surgery waiting times may reflect the possibility that waiting (for those particular specialties) causes spending. In other words, those provinces with longer waiting times (for those particular procedures) receive greater per capita government spending. This possibility was, however, thoroughly analytically explored, and rejected in two related ways. Neither the Hausman test (see Berndt, 1991) nor econometric models designed to detect reverse causation yielded significant evidence of reverse causation.

Analysis of individual provinces

While the analysis to this point has found that aggregate per capita spending has no effect on waiting time, it is possible for some individual provinces that spending does reduce waiting. One crude way to investigate this possibility is to plot, for each province, spending and waiting data (not controlling for other factors). When this is done, the provinces fall into three groups: those in which higher spending is correlated with reduced waiting, those in which higher spending is correlated with increased waiting, and those in which there is no significant correlation. The distribution of the provinces among these three groups is depicted in figures 11-13.

As illustrated in figure 11, in only two provinces, British Columbia and Quebec, was the simple correlation between spending and waiting negative and statistically significant. That is, in only those two provinces does spending, without controlling for other factors, give the impression of being wait-reducing. In five provinces - Alberta, Ontario, New Brunswick, Nova Scotia, and Prince Edward Island - the correlation between spending and waiting is insignificant (see figure 12). In particular, even though there appear to be clear patterns in the data for Ontario (negatively correlated) and New Brunswick (positively correlated), those correlations are not statistically significant. Finally, in three provinces - Saskatchewan, Manitoba, and Newfoundland - waiting and spending are significantly positively correlated (see figure 13).

Further exploration of the data for British Columbia and Quebec was undertaken to discern if the significant negative correlation between spending and waiting remained after other underlying factors were taken into account. Specifically, regressions were run in which the spending, age distribution, and time trend variables were included for each of the two provinces. This revealed that once the age distribution and the time trend were included, the relationship between spending and waiting was no longer significant for British Columbia (see table A-19). For Quebec, however, even controlling for the age distribution and the time trend, the negative and significant correlation between spending and waiting was not removed (see table A-20).19

[Previous] [Contents] [Next]

info@fraserinstitute.ca You can contact us at the above email address for any comments or information requests. Please report any dead links or technical problems.

If you know someone who would be interested in this web page, please enter their email address below, and we will forward this URL to them: Email Address:

Last Modified: Wednesday June 9, 1999.